Modeling and Analysis of Tunisia's Productive System...32 2.3.2 Global Demand Trend for Tunisian...

AfricanDevelopment Bank

(AfDB)

Tunisian Institute of Competitiveness andQuantitative Studies

(ITCEQ)

Economic Developmentand International

Finance Research Centre (DEFI)

Mr. Vincent Castel, Principal Program Coordinator, ORNA.

Mrs Natsuko Obayashi,Principal Country Economist,ORNA.

Mr. Hatem Haj Salem, Senior Operation Assistant,ORNA.

Mrs Kaouther AbderrahimBen Salah, Economist, ORNA.

Mr. Abdel Majid Ben Khlifa,Chief Economist, ModellingDepartment.

Mr. Ammar Saleheddine, IT Expert, Department of Information and KnowledgeSystem.

Mrs Mounira Bouali, Principal Economist, Department of Economic Studies.

Mrs Raoudha Hadhri, Principal Economist, Competitiveness Department.

Mrs Ikram Nahdi, Principal Economist, Modelling Department.

Mrs Samiha Chaabani, Principal Economist, Competitiveness Department.

Mrs Manel Gaaloul, Principal Economist, Department of Economic Studies.

Mr. Gilles Nancy, Professor, Aix-Marseille University.

Mr. Marcel Aloy, Maître de Conférences, Aix-Marseille University.

Mr. Eric Heyer, Professor, Sciences Po Paris.

Mr. Gilbert Cette, Associate Professor, Aix-Marseille University.

Mrs Marion Dovis, Maître de Conférences, Aix-Marseille University.

Mrs Patricia Augier, Maître de Conférences, Aix-Marseille University.

Mr. Pedro AlbuquerqueAssociate Professor, Minesota University.

Mr. Mikael Gaziorek, Professor, Sussex University.

4

T U N I S I A N I N S T I T U T E O F C O M P E T I T I V E N E S S A N D Q U A N T I T A T I V E S T U D I E S

Acknowledgements

This report was prepared based on a study on the Tunisian

productive system. It was funded by a grant from the Fund for

Middle Income Countries of the African Development Bank (AfDB).

This study is part of a capacity building of the Tunisian Institute for

Competitiveness and Quantitative Studies (ITCEQ) and is the result

of a close collaboration between ITCEQ's research team and experts

from the Centre for Research in Development and International

Finance (DEFI) of the University of Mediterranean Aix-Marseille II.

The scope of the study has covered four selected themes which

are: (i) econometric evaluation of the production factors (added value)

and prices by manufacturing sectors, from the perspective of the

production system, (ii) the econometric evaluation of the demand

stream for Tunisian export products and services by sector in

the European market, taking into account both the European

manufactures and global suppliers. Due to the lack of sufficient data,

these sectorial econometric assessments have been completed by

a panel analysis (group of sectors), (iii) Analysis of the positioning of

Tunisian exports by product on worldwide markets from the

demand/supply perspective, as well as its evolution based on

standard key performance indicators. The objective of this analysis

is to identify new product/market mixes and increase export

opportunities, (iv) Analysis of the Tunisian productive system from

the perspective of business dynamics, to determine the composition

of the productivity gains in for the labor factor (aggregate level) in

several sources: internal business and / or across several phenomena

reallocation of resources. While the themes covered by this report

are different from different angles (methodological, nature, type of

database used and their objectives), they provide complementary

perspective on the Tunisian productive system. The ultimate objective

is to identify and implement the necessary reforms to improve

economic efficiency, further diversify the Tunisian economy while

seizing the opportunities offered by globalization and technological

development and gain access to new levels of stable and inclusive

growth.

The report also includes the description of the training activities

carried out under the capacity building program for the ITCEQ

team.

Thanks are also due to ITCEQ & DEFI researchers who contributed

to the following chapters:

Chapter I:

- DEFI: Mr. Gilles Nancy, Mr. Marcel Aloy and Mr. Eric Heyer

- ITCEQ: Mr. Abdelmajid be Khalifa and Mrs. Ikram Nahdi

Chapter II:

- DEFI: Mr. Gilles Nancy, Mr. Marcel Aloy, Mr. Eric Heyer and

Mr. Pedro Albuquerque

- ITCEQ: Mrs Raoudha Hadhri and Mrs. Samiha Chaabani

Chapter III:

- DEFI: Mr. Mikael Gaziorek

- ITCEQ: Mrs. Raoudha Hadhri and Mrs. Samiha Chaabani

Chapter IV:

- DEFI: Mrs. Patricia Augier and Mrs. Marion Dovis

- ITCEQ: Mrs Mounira Bouali, Mrs. Manel Gaaloul, Mr. Abdallah

Abdelmalek & Mr. Slaheddine Ammar

The findings and opinions expressed in this report are the sole

responsibility of the authors and should not be cited without permission.

They do not necessarily reflect the views of ITCEQ or AfDB.

6


Table of Contents

7 Executive Summary

7 1. Introduction

9 2. Analysis by Component

9 2.1 Sector Mondeling of Manufacturing Industries

9 2.1.1 Estimation Method

9 2.1.2 Key Outputs

11 2.1.3 Employment Response to a Demand and Cost Shock

13 2.1.4 Capital Response to a Demand and Cost Shock

15 2.1.5 Conclusions

15 2.1.6 Bibliography

15 2.2 Sector Mondeling of Tunisian Eport’s Market Shares

16 2.2.1 Estimation of Export Functions Over the Period 1988-2008

25 2.2.2 Dynamic Panel Data Estimation of Export Equations

27 2.2.3 Possible Extensions of Econometric Analysis

30 2.2.4 Bibliography

30 2.3 Analysis of Demand for Tunisia’s Exports

30 2.3.1 Analysis of Tunisia’s Supply in terms of Comparative Advantage

32 2.3.2 Global Demand Trend for Tunisian Exports

37 2.4 Impacts of Opening up Tunisia’s Economy on the Productive System and Analysis of the Business Adaptation Process

38 2.4.1 Descriptive Analysis of Tunisia’s Industrial Business Database

46 2.4.2 Productivity Decomposition Analysis

50 3. Training

50 3.1 Example of Econometric and Macroeconnomic Modeling Training Session

51 3.2 Exemple of Training on Business Data Processing

54 4. Conclusions

7


Executive Summary

1. Introduction

The revolution disrupted the institutional framework of the Tunisian

economy. Changes in governance and transparency in public

policies will improve the efficiency of the economy, particularly the

potential growth rate. Nevertheless, the structural characteristics

of the economy and the challenges it faces remain topical.

Due to the very high concentration of textiles and apparel exports,

the dismantling of the Multi-fibre Agreement (MFA) was a major

event for the Tunisian economy. Since the collapse of MFA, Tunisia

has faced fierce competition on EU markets from Asia and Eastern

Europe, with lower labour costs and/or higher productivity. The risk

of a structural crisis in this sector is especially high, given the degree

of dependence vis-à-vis the EU which absorbs 96% of textiles and

apparel exports from Tunisia. Furthermore, most of these exports

come from outsourcing, a vulnerable low- value-added business

activity.

However, there are some bright spots in the current structure of

manufactured exports. There are a number of new, fast-growing

export commodities, such as cable assemblies, electronic components,

plastics, essential oils and detergents. Their export share is, however,

very minimal.

Improvement of competitiveness is at the heart of a development

project based on integration into the global economy. Beyond

the indispensable modernization of technological processes, the

functioning of the labour market and specifically the wage bargaining

process, de facto initiated and supervised by the State, is not

sufficiently flexible to respond to competition from other emerging

countries on major markets. Failing to tailor the training system to the

Tunisian production model, and consequently to the demand from

businesses, hinders the improvement of competitiveness.

The share of services in value added is growing. The liberalization

of services offering speedy export opportunities should primarily

focus on promising and innovative sectors with high value added

such as IT services, engineering, accounting, auditing and

management consulting, publishing and editing, educational

services, management of public services and health services. The

implementation of these reforms would, in the medium term,

significantlyreduce Tunisia's vulnerability to fluctuations in demand

for tourism services.

Another pillar of the Tunisian development model is the growth of

FDI, which contributes to financing the needs of the economy,

despite its adverse effects on the current account (income transfers

abroad), and participates in the rapid increase in the rate of

investment.

Numerous analyses and discussions have been initiated to meet

the challenges facing the Tunisian economy. Competitiveness,

dynamics of the manufacturing sector, impact of external shocks,

role of government in an economy open to the outside world and

a better understanding of potential GDP underpin the programme

of the Technical Assistance Mission. Naturally, such a short-term

mission could not presumably address all these issues. DEFI and

ITCEQ experts have endeavoured to delimit the scope of objectives

and activities to be developed under the project, given the duration of

the mission, contingent events, and the availability of information

necessary for the completion of the programme.

8


It is against such backdrop that the DEFI/ITCEQ Technical Assis-

tance mission should be repositioned. Indeed, the programme ad-

dresses, without dealing with them exhaustively, the recurrent

issues facing the Tunisian economy such as productivity of factors,

export competitiveness and substitution determinants between

non-graduates and graduates on the labour market.

The Tunisian Institute of Competitiveness and Quantitative Studies

(ITCEQ) is a public non-administrative establishment under the

supervisory authority of the Tunisian Ministry of Development and

International Cooperation. It plays a key role in planning and

programming the economic policy of the Government of Tunisia, as

well as conducting analyses necessary for public decision-making.

Within the context of Tunisia's integration into the global economy,

ITCEQ sought to strengthen its technical capacity with respect to

analyses required for public decision-making and to contribute to

the economic policy decisions of the Tunisian Government.

DEFI (Economic Development and International Finance Research

Centre) of the University of the Mediterranean (Université de la

Méditerranée), specialized in analysing the mechanisms for the

integration of emerging countries into the world economy, was

requested to implement a technical assistance mission under the

aegis of the ADB, in close conjunction with ITCEQ researchers,

whose goals, methods and findings are the subject of this

report.

1 After taking into account all the net financial flows, excluding post-crisis remittances and official loans.2 Assuming a contribution of USD 900 million under this programme (WB, AfDB, EU, AFD).

9


2. Analysis by Component

2.1 Sector Modelling of Manufacturing Industries

The modelling of manufacturing industries rests on theoretical

foundations, developed in the appendices and structured

after general equilibrium models by incorporating therein elements

of demand. The objectives of modelling are to:

- Simulate and forecast the value-added growth rate of

manufacturing sub-sectors based on structural equations;

- Estimate the demand for capital input, labour for each sub-

sector;

- Estimate the elasticity of factor substitution and, for the labour

factor, the elasticity of substitution between skilled and unskilled

labour;

- Determine the price of value-added in the manufacturing sector;

- Distribute supply and demand between the domestic market

and foreign markets; and

- Provide the factor (employment and investment) demand

response to a demand (growth in value-added) and cost (salary)

shock.

The estimation period spans from 1983 to 2009. Data was collected

from the annual national accounts for the Tunisian economy. Six

manufacturing sectors are analysed, namely:

Sector 2: Agricultural and Food Industry (IAA);

Sector 3: Ceramic Building Materials and Glass Industries

(MCCV);

Sector 4: Mechanical and Electrical Industries (EMI);

Sector 5: Chemical Industries (CHEMI);

Sector 6: Textile Apparel and Leather Industries (THC); and

Sector 7: Miscellaneous Industries (MISCEL).

2.1.1. Estimation Method

The low number of observations available (at most 27) and the presence

of common parameters to be estimated in the various long-term

equations (elasticities of substitution, the growth rate of technological

progress) prompted the simultaneous estimation of these equations.

Consequently, the following two-stage process was adopted:

1. In stage one, long-term relationships were estimated, per level,

through a simultaneous equations system using the SUR

(Seemingly Unrelated Regression) method.

2. In stage two, 4 ECMs (Error Correction Models) were estimated

by imposing therein relations estimated in stage one as

long-term solution.

All ECMs have satisfactory statistical properties. LM tests result in

the rejection of the hypothesis of autocorrelation in the residuals of

these equations. These residuals are homoskedastic under the White

test and the ARCH test. The functional form of the equation passed

the Reset test. Lastly, according to the Jarque Bera test, the residuals

of all equations are normally distributed.

2.1.2. Key Outputs

The key outputs are summarized in Table 1: The following conclusions

ensue:

• There would be a relatively high substitutability between capital

and labour in 4 sectors: the elasticity of substitution is close to

unity in the Agri-food industry (IAA) as well as in the Mechanical

10


and Electrical industry (EMI) (sectors 2 and 4). The elasticity of

substitution is close to 0.7 in the Ceramic Building Materials and

Glass and Chemicals industries (sectors 3 and 5). However, it is

worth noting that the value of these elasticities is probably over-

estimated by including working hours in labour. In other words,

this does not involve elasticity between capital and labour, which

is being estimated here, but between capital and work hours.

In other words, an increase in the relative price of wages

compared to the cost of capital would lead to high substitution

of capital for labour in the IAA and EMI. However, in other sectors,

the change in the relative price in both sectors has a lower but

significant incidence on the respective demand for labour and

capital in both sectors.

• However, there seems to be a strong complementarity between

capital and labour in the last 2 sectors under review, namely

Textile, Apparel and Leather and miscellaneous industries (sectors

6 and 7). This implies that in these sectors, and unlike the previous

ones, a variation in the relative cost of labour relative to capital

will have little long-term impact on capital intensity (capital stock

per worker).

• As per our estimates, it seems that the formulation of a Hicks-3

neutral technological progress is the only one to be accepted in

the six sectors studied. In sectors 2, 6 and 7, the estimated

growth rate of technical progress (gk and gl parameters, which

are identical in the Hicks-neutral technology parameters) is

between 1 and 2% per year. It is close to 2.7% in sector 4 and

above 5% in sectors 3 and 5.

The growth rate of global factor productivity is heterogeneous. It is

relatively low in IAA and textiles and high in EMI and the chemical

industry.

• In all the sectors studied, there is a high elasticity of substitution

between skilled and unskilled employment ranging from 3.3 for

sector 5 to more than 6 for sector 4. This outcome is significant,

apparently robust and relatively unexpected: It implies in particular

that a 1% decrease in the relative wage of skilled workers versus

unskilled workers would lead (in the long run) to an increase in

the number of jobs for skilled workers higher than that of unskilled

workers by 3% to 6%. In other words, a tighter wage gap

between skilled and unskilled workers would be, according to

this estimate, an effective means of improving the "employability"

of graduates. One possible interpretation of this result is that,

on average, graduates would not have enough specific expertise

that would suitably distinguish them from unskilled workers, given

that both categories are consequently considered by businesses

as substitutes rather than complements in the production

process.

• In conclusion, it should be recalled that long-term elasticities

with respect to quantities were fixed to unity, for obvious reasons

of theoretical consistency: thus, when the total amount of work

increases by 1%, the amount of hours put in by skilled and

unskilled workers increases identically by 1% (all things being

equal). Similarly, when demand for goods (measured by value

added in volume) increases by 1%, the amount of capital and

labour increases identically by 1%.

3 Technological progress is Hicks-neutral if it increases the effectiveness of both factors of production. For a given productive combination, the ratio of marginal productivitiesremains unchanged and the demand for factors increases to keep pace with technical progress.4 It should be noted that the distinction between skilled/unskilled workers here is understood in terms of degree and not the type of business: in our estimates, skilled workerswere defined as higher education graduates, while the unskilled comprise other individuals. The terms used in our presentation ("skilled workers" and "unskilled workers") aretherefore deliberately simplistic and highly schematic. To complete the estimation, long-term equations were placed in the broader context of error correction models. The genericform of the dynamic equations estimated is given in Table III.4. As above, the non-significant short-term coefficients were removed from the final equations.

11


To illustrate the dynamic process of the different variables and their

sensitivity to economic determinants, the assessment of sector

responses to some economic shocks was conducted.

2.1.3. Employment Response to Demand and Cost Shock

Two shocks were simulated: the first is a 1% increase in value added,

whereas the second shock is a 1% increase in real wages.

The outcomes of these simulations are summarized in Table 2 and

in the graphs below. Responses to shocks reveal that in the wake

of a shock, employment does not recover its initial value. In the

case of a demand shock (VA), the employment response is fastest

in the MCCV, EMI and THC sectors. The shocks are still very

persistent 4 years later. Following an increase in real wages, the

adverse effects on employment are felt instantly in the IAA and to

a lesser extent in EMI and the chemicals industry. In the long run,

the IAA and EMI are the sectors whose jobs are the most affected.

In contrast, employment in the textile sector seems to better absorb

the shock and be relatively preserved in the long run.

Table 1: Trend of Economic Indicators (2008-2011)

Sources: National Accounting, ITCEQ/DEFI calculations

Sector 2 3 4 5 6 5

Structural Parameters

Elasticity of substitution K/L 0.99 0.74 0.98 0.67 0.06 0.14

Elasticity of substitution Skilled/Unskilled labour

5.40 4.10 6.01 3.29 4.67 5.55

Hicks-Neutral technical progress(%)

1.97 5.11 2.68 5.28 1.06 1.67

ECM Parameters

Employment buoyancy 0.40 0.21 0.32 0.12 0.21 0.31

Capital buoyancy 0.05 0.13 0.07 0.06 0.14 0.04

Table 2: Dynamic Response of a 1% VA and Real Wages (RW) Shock on Employment)


Sector T 1 year 2 years 3 years 4 years Long term

IAA (2)VA 0.31 0.46 0.62 0.71 0.79 1

RW -1.16 -0.63 -0.97 -0.84 -0.94 -0.98

MCCV(3)VA 0.85 0.88 0.38 0.52 0.62 1

RW 0 -0.77 -0.76 -0.75 0.74 -0.73

IME(4)VA 0.44 0.62 0.74 0.82 0.88 1

RW -0.21 -0.46 -0.62 -0.73 -0.81 -0.97

CHEMI (5)VA 0.19 0.30 0.28 0.31 0.32 1

RW -0.22 -0.06 -0.01 -0.20 -0.28 -0.67

THC(6)VA 0.37 0.50 0.61 0.69 0.76 1

RW -0.47 -0.38 -0.31 -0.26 -0.21 -0.05

MISCEL. (7)VA 0.53 0.68 0.78 0.85 0.90 1

RW 0 -0.04 -0.07 -0,09 -0.11 -0.13

12


Graph 1: Employment variation (in %) following a 1% increase in value added

Graph 2: Employment variation (in %) following a 1% increase in real wages

13

2.1.4. Capital Response to Demand and Cost Shock

Two shocks were also simulated for capital:

1. A 1% increase in value added

2. A 1% increase in the real cost of capital

The results of these simulations are summarized in Table 3 and in

the graphs below. Responses to capital shocks show a large enough

inertia in the short term in most sectors, except textiles, for value

added. In contrast, the long-term effects of an increase in the real

cost of capital5 are substantial in IAA, IME, and MCCV, but negligible

in the textile and apparel sector.


Table 3: Dynamic Response of a 1% VA Shock and Actual Cost (CKR) on Capital Stock


Sector T 1 year 2 years 3 years 4 years Long term

IAA

VA 0.18 0.51 0.71 0.83 0.91 1

RW -0.10 0.27 -0.40 -0.50 -0.64 -0.98

MCCV

VA 0.17 0.36 0.58 0.78 0.93 1

RW -0.14 -0.29 -0.45 -0.61 0.73 -0.73

IME

VA 0 0.18 0.42 0.65 0.83 1

RW 0 -0.07 -0.21 -0.39 -0.56 -0.97

CHEMI

VA 0 0.15 0.25 0.39 0.52 1

RW -0.03 -0.15 -0.24 -0.34 -0.43 -0.67

THC

VA 0.27 0.50 0.67 0.79 0.87 1

RW -0.008 -0.028 -0.036 0.042 -0.047 -0.05

MISCEL.

VA 0 0.04 0.10 0.17 0.24 1

RW -0.04 -0.07 -0.08 -0,09 -0.10 -0.13

5 Real cost of capital: the real interest rate plus the rate of capital depreciation.

14


Graph 3: Capital stock variation (in %) following a 1% increase in value added

40

Graph 4: Employment variation (in %) following a 1 % increase in actual cost

42

15


2.1.5. Conclusions

The modelling of micro-manufacturing sectors, based on approaches

to computable general equilibrium models, has given rise to estimates

of parameters leading to the following conclusions:

- The substitution between capital and labour is high (elasticity

close to unity, Cobb-Douglas) in the Food and Agricultural Industry

as well as in the Mechanical and Electrical Industries (sectors 2

and 4), both of which are strong exporters. Substitution between

both factors is weaker (elasticity close to 0.7, CES) in the MCCV

and Chemicals sectors. In other sectors, there is complementarity

between production factors;

- Dynamic simulations reveal that Tunisia's industrial sectors are

characterized by a relatively high rigidity with respect to the

adjustment of factors: in the wake of demand or actual cost

shocks, it generally takes three to four years for a significant

adjustment to occur in the quantity of labour and capital;

- The global productivity growth rate of factors is heterogeneous.

It is relatively low in IAA and textiles, and high in EMI and the

chemicals industry;

- In all the sectors studied, there is a high elasticity of substitution

between skilled and unskilled employment6 ranging from 3.3 for






workers by 3% to 6%.

Consequently, these initial findings allow for the validation of both

the theoretical and econometric methodologies implemented. Hence,

based on the work done in this study, sector estimates of prices and

foreign trade - which still have to be done - must be conducted

without major difficulties.

2.1.6. Bibliography

Annabi N., Cockburn J., Decaluwé B. (2003), Formes

Fonctionnelles et Paramétrisation dans les MCEG, CREFA,

Université de Laval

Harrison R., Nikolov K., Quinn M., Ramsay G., Scott A. and

Thomas R. (2005), The Bank of England Quarterly Model,

www.bankofengland.co.uk/publications/beqm/

Klump R., McAdam P., Willman A. (2008), Unwrapping some euro

area growth puzzles: Factor substitution, productivity and

unemployment. Journal of Macroeconomics 30, 645–666

Lofgren H., Harris R.L., Robinson S. (2002), A standard computable

general equilibrium model in GAMS. Microcomputer in Policy

Research 5, International Food Policy Research Institute.

2.2 Sector Modelling of Tunisian Exports MarketShares

The analysis of the determinants of Tunisia's exports covers eight

sectors identified by ITCEQ experts, as follows:

Sector 1: Agriculture (Agr)

Sector 2: Agri-food Industry (IAA)

Sector 3: Ceramic Building Materials and Glass (MCCV)

Sector 4: Mechanical and Electrical Industries (IME)

Sector 5: Chemical Industries (CHEMI)

Sector 6: Textile, Apparel and Leather Industries (THC)

6 It should be noted that the distinction between skilled/unskilled workers here is understood in terms of degree and not the type of business: in our estimates, skilled workerswere defined as higher education graduates, while the unskilled comprise other individuals. The terms used in our presentation ("skilled workers" and "unskilled workers") aretherefore deliberately simplistic and highly schematic.

16


Sector 7: Miscellaneous Industries (MISCEL)

Sector 8: Hydrocarbons (Hyd)

The objective is to conduct a quantitative assessment of sector

exports dynamics as a result of variations in expressed demand and

relative prices.

Under the project, implementation is limited to the study of Tunisia's

exports toward Europe, which represents the bulk of its exports,

based on annual data covering the period 1988-2008.

i) The explanatory variables of Europe’s7 import demand are:

- European expressed demand;

- Tunisian sector export price index;

- European price index; and

- Tunisia's competitor price index on the European market.

ii) The theoretical modelling included in the Annex is an extension

of conventional export models8.

iii) The estimation of econometric equations is based on the

specification of error-correction models. This approach has the

advantage of distinguishing between short-term elasticities and

long-term elasticities of European sector imports.

iv) This methodology is supplemented by dynamic panel estimates

where the sample size is much larger and therefore allows for

the testing of the robustness of the results of time series data

comprising 21 observations covering the period 1988-2008.

v) In addition, some interesting extensions in terms of modelling,

that can be undertaken as a continuation of existing work, are

suggested and initiated:

- First, econometric estimates may be supplemented by a sector

model for the modelling of prices and quantities (imports and

exports), leading to the assessment of nominal flows and trade

balances by sector;

- Second, modelling may introduce nonlinearities. These take into

account the possibility of time-varying elasticity. For example,

price elasticities may be lower when price differentials are small

and more significant when the price differentials increase, due

to the existence of transaction costs; and

2.2.1. Estimation of export functions on time series datacovering the period 1988-2008

2.2.2.1. The theoretical modelling reported in the Annex proposes

three potential explanatory variables for Tunisian exports:


- Price index of Tunisian sector export models;



However, preliminary estimates of export functions by sector

revealed collinearity problems between different price indexes that

made it difficult to interpret the estimated coefficients. Therefore,

the following dynamic specification was retained, which is restricted

to the consideration of only one relative price (p) which, as stated

below, may take two different specifications:

(1) Δx(t) = ρΔx(t-1) + α0Δd(t) + α1Δd(t-1) + α2 Δd(t-2) + φ0Δp(t)

+ φ1Δp(t-1) + φ2Δp(t-2)

+ γ [ x(t-1) - βp(t-1) – d(t-1) + μ.t ] + εt

x: logarithm of Tunisian exports to the European Union (EU) in

thousand Euros constant base 100 = 2005 (compiled from Comext).

7 See for example Annabi N., Cockburn J., Decaluwé B. (2003) for a presentation on the microeconomic foundations of demand functions.8 See for example Wong (2008) for a recent application to the case of Malaysia.

17

d: logarithm of the demand made to Tunisia by the EU at constant

prices (compiled from Comext). The demand will be expressed in 2

alternative ways:

- The first considers demand expressed for all commodities in the

sector concerned (Agriculture, IAA, ...)

- The second limits the demand to Tunisia's main exports in a

given sector (Agriculture, IAA,). Indeed, there is considerable

heterogeneity in the composition of commodities from each

sector, and in this form, expressed demand may be the more

representative model for Tunisian exports.

p: logarithm of the relative price of Tunisian exports. This relative

price is expressed in 2 alternative ways:

- The first one retains an index of sector relative prices for Tunisian

exports with respect to EU imports in the same sector;

- The second uses an index of relative prices for Tunisian exports

with respect to those of major competitors on the European market.

Hence, for each sector, 4 formulations were estimated according to

whether expressed demand should be calculated on all commodities

or restricted to the major export, and whether the relative price

should be calculated based on competitor prices or that of European

Union countries.

In this error-correction model (ECM), the term in brackets [ x(t-1) - βp(t-

1) – d(t-1)] represents the deviation from long-term equilibrium, wherein

long-run elasticities are equal to 1 with respect to expressed demand

(in fact, Tunisia’s market share trend is modelled on the European

market) and from β with respect to the relative price.

The term μ.t stands for the potential deterministic trend of market

shares, with an annual growth rate μ.

Lastly, the variables Δx represent the variation of x, and

dummy variables were sometimes included in the regression

model.

2.2.2.2. Key Outputs

1) Agriculture

The main findings, for the 4 specifications tested for this sector,

are summarized in Table 4:

• The value of the long-run elasticity of demand for exports was

imposed to unity. It is worth noting that, in preliminary tests, this

restriction is more easily accepted by using expressed demand

calculated on the basis of major commodities;

• Whenever expressed demand for "Total exports" is used, our

estimates reveal a significantly positive trend, representing an

increase in market shares trend for this sector;

• Long-term price elasticities bear the expected sign in three out

of four cases. Only the formulation with EU prices and "major

exports" for demand bears a positive sign, contrary to economic

intuition. The value of this elasticity is low, to the tune of-.2 in

the case of competitor prices and -0.3 in that of EU countries,

which would imply that Tunisian products in the sector have few

substitutes on the EU import market; and

• ECMs that use "Total commodities" for demand have satisfactory

statistical properties. LM tests result in the rejection of the

hypothesis of autocorrelation in the residuals of these equations.

These residuals are homoskedastic under the White test and

the ARCH test. The functional form of the equation passed the

Reset test. Lastly, according to the Jarque Bera test, the residuals

of all equations are normally distributed. That is not the case for

hose using "Major exports" for demand.


18


Table 4: Summary of Key Findings of the Export Equation Estimates for Sector 1

Source: ITCEQ/DEFI Calculations / Note: Student's t-statistic is shown in bracketsne: coefficient was not estimated but imposed.

α0 Total Commodities Major Commodities

Expressed Demand 1ne 1ne 1ne 1ne

Buoyancy -0.63 (-3.70) -0.31 (-1.45) -0.38 (-2.21) -0.65 (-3.28) γ

Relative Prices of …… competitors -0.20 (-0.28) -0.17 (-1.00) β

… EU as a whole -0.30 (-2.09) 0.06 (0.12)

Trend 0.007 (2.12) 0.02 (1.81) μ

ST ElasticitiesΔlog(expressed demand) 1.49 (3.57) α0Dummyd98 0.16** (1.91) 0.16* (2.03) η0d06 0.23 (2.32) η1d02 -0.14 (-1.72) η2TestsLM (2) 0.82 0.21 0.03 0.10

Arch(1) 0.72 0.33 0.72 0.33

Normality 0.71 0.56 0.92 0.38

Reset (2) 0.97 0.29 0.08 0.05

R-squared 0.63 0.50 0.44 0.39

Adjusted R-squared 0.46 0.32 0.34 0.33

Ranking 1 4 3 2

At the end of these estimates, it seems the best equation is one that

uses "Total exports" as expressed demand and considers the prices

of EU countries as foreign price.

2) IAA Sector

• The value of elasticity of demand for exports was imposed to

unity. It is worth noting that this restriction is accepted in all

cases;

• Long-term price elasticities have the expected sign in three

out of four cases. Only the formulation with the price of EU

countries and "major exports" for demand bears a positive

sign, contrary to economic intuition. The value of this elasticity

is -0.7 in the case of competitor prices and -0.5 in that of EU

countries; and

• All ECMs have satisfactory statistical properties. LM tests result

in the rejection of the hypothesis of autocorrelation in the

residuals of these equations. These residuals are homoskedastic

under the White test and the ARCH test. The functional form of

the equation passed the Reset test. Lastly, according to the

Jarque Bera test, the residuals of all equations are normally

distributed.

According to these estimates, the best equation is the one that uses

"Total exports" for demand and considers the prices of EU countries

as foreign price. Findings are summarized in the table below.

19


Table 5: Summary of Key Findings of the Export Equation Estimates for Sector 2 Estimation (Period: 1988-2008)


Total Commodities Major Commodities

LT Elasticities

Expressed demand 1ne 1ne 1ne 1ne

Buoyancy -1.13 (-6.10) -1.19 (-5.82) -1.19 (-9.94) -1.33 (-5.99)γ

Relative Prices of …

… competitors -0.64 (-1.87) -0.74 (-2.39)β

… EU as a whole -0.54 (-1.61) 0.26 (2.47)

Trend μ

ST Elasticities

Δ( Relative prices)-0.89 (-2.61) -0.93 (-2.54) -0.75* (-2.06)

φ0

Dummy

d02 -0.93 (-3.52) -0.87 (-3.25) -0.98 (-3.97) -1.12 (-4.11)η0

d01 -0.96 (-3.10) -0.90 (-2.74) -0.93 (-3.13) -1.15 (-3.32)η1

d96 -0.79 (-3.32)η2

d94 0.48* (2.09) 0.62 (3.74)η3

Tests

LM (2) 0.55 0.32 0.37 0.87

Arch(1) 0.85 0.70 0.33 0.93

Normality 0.80 0.86 0.83 0.74

Reset (2) 0.10 0.52 0.12 0.76

R-squared 0.87 0.81 0.86 0.82


Ranking 1 3 4 2

3) Ceramic Building Materials and Glass Sector (MCCV)

• The value of elasticity of exports demand was imposed to unity.

It is worth noting that this restriction is accepted econometrically

in the case of "Total commodities". The elasticity in the case

of "Major exports" is close to 0.6 when the coefficient is

left free.

• Long-term price elasticities bear the expected sign in all cases.

he value of this elasticity is close to -0.5, indicating a significant

price effect.

• ECMs that use "Total exports" for demand have satisfactory

statistical properties. LM tests result in the rejection of the hypothesis

of autocorrelation in the residuals of these equations. These

residuals are homoskedastic under the White test and the ARCH

test. The functional form of the equation passed the Reset test.

Lastly, according to the Jarque Bera test, the residuals of all

equations are normally distributed. That is not the case for those

using "Major exports" for demand.

20


Table 6: Summary of Key Findings of the Export Equation Estimates for Sector 3 (Period: 1988-2008)



LT Elasticities


Buoyancy -0.88 (-6.66) -0.87 (-6.63) -0.74 (-4.70) -0.73 (-4.92) γ


… competitors -0.46 (-3.81) -0.57 (-3.21) β

… EU as a whole -0.44 (-3.64) -0.49 (-2.69)

Trend μ

ST Elasticities

Dummy

d9293 -0.52 (-2.86) -0.51 (-2.84) -0.68 (-2.78) -0.63 (-2.76) η0

Tests

LM (2) 0.52 0.70 0.49 0.61

Arch(1) 0.35 0.41 0.58 0.75

Normality 0.63 0.68 0.26 0.23

Reset (2) 0.37 0.23 0.01 0.01

R-squared 0.85 0.86 0.76 0.79


Ranking 1bis 1 3bis 3

4) Mechanical and Electrical Industries Sector


unity. Unstressed, such elasticity would exceed unity, reaching

around 1.3 in both cases.

• In all ECMs, estimates reveal a significantly positive trend,

signifying an increased market share trend in this sector. This

is consistent with elasticity of expressed demand exceeding

unity.

• Long-term price elasticities do not have the expected sign when

competitor prices are included in the relative price. In the other

case, the estimated elasticity is relatively high, between -

1 and -1.5.




under the White test and the ARCH test. The functional form

of the equation passed the Reset test. Lastly, according to the


distributed.

This sector over-reacts to the European economic situation and

is also very competitive.

21




LT Elasticities


Buoyancy -0.22 (-2.31) -0.13 (-2.27) -0.66 (-8.86) -0.73 (-2.71) γ


… competitors 0.09 0.27 β

… EU as a whole -1.10 -1.56

Trend 0.09 0.08 0.10 0.06 μ

ST Elasticities

Δlog(Expressed Demand) 0.42 (4.36) 0.35 (2.92) 0.72 (8.91) α0

Δ( Relative Prices) -0.32 (-2.77) -0.14 (-1.24) -0.53 (-6.24) φ0

Δ( Relative Prices) -1 0.39 (4.12) φ1

Δlog(Exports)-2 -0.28 (-4.91) ρ0

Dummy

d89 0.14 (3.60) 0.12 (2.48) η0

d06 0.16 (3.28) 0.17 (2.70) 0.10 (7.65) η1

d96 -0.18 (-3.68) η2

Tests

LM (2) 0.34 0.54 0.36 0.61

Arch(1) 0.72 0.86 0.16 0.32

Normality 0.67 0.34 0.72 0.35

Reset (2) 0.64 0.40 0.83 0.35

R-squared 0.83 0.75 0.98 0.63


Ranking 1bis 3 1 3bis


The best equation is the one that uses the prices of all EU coun-

tries in the competitiveness indicator, and for major commodities

only.

5) Chemical Sector



around 1.5 in all cases, except for case 4 (expressed demand,

“major commodities”, competitor prices), where elasticity stands

at 0.8;

• In all ECMs, estimates reveal a significantly negative trend,

signifying a loss-of-market share trend in this sector. This is

consistent with elasticity of demand below unity;

• Price elasticities bear the expected sign, regardless of the

formulation used, although their value varies greatly: it stands

around -0.7 / -0.9 in the case of a "Total commodities" demand.

These values become much higher (-1.89 and -4.67 respectively)

when expressed demand includes "Major commodities". This

indicates a high degree of substitutability of Tunisia's chemical

exports with respect to competing exports;




under the White test and the ARCH test. Lastly, according to

the Jarque Bera test, the residuals of all equations are normally

distributed. The functional form of the equation alone failed the

Reset test.

22





LT Elasticities


Buoyancy -0.75 (-2.70) -0.67 (-2.36) -0.23 (-1.99) -0.39 (-3.55) γ


… competitors -0.88 (-2.35) -4.67 (-3.91) β

… EU as a whole -0.70 (-2.69) -1.89 (-2.09)

Trend -0.06 (-3.40) -0.06 (-3.19) -0.11 (-2.63) -0.13 (-4.47) μ

ST Elasticities

Δlog(Expressed Demand) α0

Δ( Relative Prices) -0.15 ** (-1.79) α1

Δ( Relative Prices) -1 1.52 (2.90) φ1

Δlog(Exports)-2 1.39 3.00) ρ2

Dummy

d9394 -0.18* (-2.26) -0.19 * (-2.28) -0.24 (-2.69) -0.28 (-4.08) η0

Tests

LM (2) 0.75 0.42 0.73 0.19

Arch(1) 0.49 0.91 0.37 0.30

Normality 0.63 0.68 0.61 0.42

Reset (2) 0.07 0.10 0.07 0.02

R-squared 0.74 0.73 0.65 0.86


According to these estimates, it would seem the best equation is

one that uses prices calculated based on competitor prices in the

competitiveness indicator and "Major exports" for demand.

6) Textile, Apparel and Leather Sector


unity. Such unstressed elasticity would fall below unity, and

stand at about 0.6 in all cases;


formulation used, and their value is high, stabilized between -

3 and -4 when competitors’ prices are used and between -

5 and -8 in the other case. Consequently, as in the case of

chemical industries, there would be a phenomenon of high

substitutability of these exports on the European market.







distributed.

23





LT Elasticities


Buoyancy -0.14 (-2.12) -0.15 (-3.31) -0.10 (-2.21) -0.15 (-3.56) γ


… competitors - 2.88 (-9.01) -4.21 (-8.27) β

… EU as a whole -5.30 (-4.99) -8.04 (-6.39)

Trend μ

ST Elasticities

Δ( Relative Prices) -0.26 (-2.33) φ0

Dummy

d01 0.16 (2.41) 0.15 (2.25) 0.19 (2.73) η0

Tests

LM (2) 0.84 0.48 0.62 0.46

Arch(1) 0.36 0.69 0.62 0.88

Normality 0.69 0.98 0.64 0.58

Reset (2) 0.16 0.89 0.77 0.77

R-squared 0.65 0.84 0.80 0.86


Ranking 4 2 3 1

According to these estimates, the best equation is one that uses

prices calculated based on competitor prices in the competitiveness

indicator and "Major exports" as demand.

7) Hydrocarbons and Refined Products Sector


unity. It is worth noting that this restriction is more easily accepted

when using expressed demand calculated on the basis of “Total

commodities” (estimated elasticity is1.03) than in the case of

“Major commodities” (elasticity estimated above 0.8);

• Price elasticities always have the expected sign. The value

ranges between -8 in the case of EU prices and around -5 in

the case of competitor prices. This sector is characterized by

the highest price elasticities among the sectors studied. This

is not surprising, given the type of products involved in the

sector: hydrocarbons are standardized exports, and Tunisian

commodities are close substitutes with respect to exports traded

on the European market.







distributed.

24





LT Elasticities


Buoyancy -0.83 (-6.31) -0.41 (-2.31) -0.66 (-3.75) -0.38 (-2.02) γ


… competitors -5.65 (-1.69) -4.47 (-1.29) β

… EU as a whole -7.93 -7.93 (-3.81)

Trend μ

ST Elasticities

Dummy

d9394 -0.53 (-4.52) -0.68 (-3.60) -0.65 (-5.09) -0.71 (-3.74) η0

Tests

LM (2) 0.11 0.26 0.65 0.30

Arch(1) 0.83 0.84 0.12 0.84

Normality 0.89 0.27 0.73 0.21

Reset (2) 0.80 0.68 0.30 0.62

R-squared 0.89 0.69 0.86 0.67


Ranking 1 3 2 4

The best equation is that which uses “Total exports” as demand

and EU prices as foreign price.

8) Estimation of Global Equation


unity. It is worth noting that this restriction is more easily

accepted when using expressed demand calculated on the

basis of “Total commodities” (estimated elasticity is1.01) than

in the case of “Major commodities” (elasticity estimated

above 0.9);

• Price elasticities always have the expected sign except for

formulation 3 (prices of all EU countries and expressed demand

(Major commodities). The value ranges between -0.3 and -0.5

in the other cases;




under the White test and the ARCH test. The functional form of

the equation passed the Reset test. Lastly, according to theJarque

Bera test, the residuals of all equations are normally distributed.

25


Table 11: Summary of Key Findings of the Export Equation Estimates for Total Goods (Period: 1988-2008)



LT Elasticities


Buoyancy -0.53 -0.49 -0.53 -0.57 γ


… competitors -0.49 -0.31 β

… EU as a whole -0.53 0.35

ST Elasticities

Δlog(Expressed Demand)-1 -0.44 α1

Δ( Relative Prices)-1 -0.35 φ1

Dummy

d9394 -0.14 -0.11 -0.13 -0.12 η0

Tests

LM (2) 0.59 0.36 0.78 0.87

Arch(1) 0.55 0.86 0.51 0.43

Normality 0.73 0.96 0.20 0.42

Reset (2) 0.08 0.17 0.14 0.87

R-squared 0.78 0.83 0.75 0.82


Ranking 3 2 4 1

The best equation is that which uses “Total exports” as demand

and competitor prices as foreign price.

2.2.2.3. Conclusion

In conclusion, the findings of the estimates conducted provide

important lessons, as they highlight huge differences among

sectors, both with respect to the long-term price elasticities of the

various sectors (and hence their level of substitutability on the

European market), and to the dynamic behaviour of Tunisian

exports in the wake of relative price shocks and demand shocks

(see Appendix). Henceforth, the estimation of an aggregate export

equation, even if it meets econometric quality criteria, cannot

underpin forecasts and the formulation of a coherent industrial

development policy.

2.2.2. Estimation of export functions on time series datacovering the period 1988-2008

Dynamic panel estimates confirm the results obtained from time

series data.

2.2.2.1. Estimation Methods

Three econometric approaches consistent with the methods used

for time series data were developed, namely:

26


i) an autoregressive distributed lag (ADL) model in levels (assuming

stationarity of the series), by using two alternative specifications

for relative prices: one based on EU prices and the other based

on competitor prices on the European market;

ii) an autoregressive distributed lag (ADL) model in first differences

(assuming non-stationarity of the series), by using the two

alternative specifications for relative prices; and

iii) an error-correction model (ECM) estimated in one step, which

allows the inclusion of integrated and non-cointegrated cases,

still under the two alternative specifications for relative prices.

2.2.2.2. Key Findings

1) ADL Model in levels in the case of EU prices (LPUE)The estimated value of the elasticity of demand for exports is 0.82

and 0.27 in the long term and short term, respectively.

The estimates reveal a significantly positive trend, which represents

an increased market share trend.

Price elasticities bear the expected sign: the value of this elasticity

is -0.71 in the short term and long term.

The ADL model has satisfactory statistical properties.

2) ADL Model in levels in the case of competitor prices (LPCON)

The value of the elasticity of demand for exports is 0.83 in the long

run and 0.26 in the short run.

Our estimates reveal a significantly positive trend, signifying an

increased market share trend in this sector.

Estimated price elasticities stand at -0.65 in the short term and -

0.71 in long term.

Residuals of this equation have satisfactory statistical properties.

There is very little difference between the two specifications, one

that includes the EU prices in the relative price and the other which

uses competitor prices on the European market.

3) ADL Model in variation

Whatever the relative prices chosen, with respect to Tunisia’s

competitors or the EU countries, the short-run elasticity of demand

by the EU is not significant. In contrast, for the two specifications

(competitor prices and EU prices), price elasticities are significant,

ranging from -0.65 in the short term to -0.7 in the long-term, and

equivalent to those obtained for the model in level.

4) ECM Model in the Case of Prices of Competitor Countries

The value of the elasticity of demand for exports is 0.83 in the long

run and 0.26 in the short run.

Our estimates reveal a significantly positive trend, signifying an

increased market share trend in this sector.

Estimated price elasticities stand at -0.65 in the short term and -

0.71 in long term.

Residuals of this equation have satisfactory statistical properties.

2.2.2.3. Conclusions

In conclusion, the data panel estimates helped to meaningfully

supplement the estimates made for each sector considered separately.

Given the low number of observations in the sample studied, data

panel estimation allows for a rather robust estimation of the 'average’

dynamic behaviour of sector exports. It follows that, whatever the

specifications used for the measurement of relative prices:

i) The elasticity of Tunisian exports to Europe in relation to expressed

demand is close to unity in the long run (this unit value is not

statistically rejected in ECMs), but, on average, seems relatively

low in the short term and even non-significant in growth rate

estimates. As a result, the instantaneous effect of a variation in

expressed demand may be considered negligible, given that

sector export adjustments probably require a time frame

substantially greater than one year.

ii) The price elasticity of Tunisian exports to Europe stands at -0.6

to -0.73 both in the short and long terms.

27

iii) Lastly, the residual plots of the various models estimated suggest

that the IAA, IMCCV (the first sub-period) and hydrocarbons

sectors are the least close to the overall estimate, which justifies

the sector approach. Indeed, the time series data reveal that the

hydrocarbons sector has much higher price elasticity than other

sectors, while the IAA sector has relatively high short-term

demand elasticities compared to other export sectors.

2.2.3. Possible Extensions of Econometric Analysis

2.2.3.1. A Foreign Trade Model

All analytical forms of the various equations are developed in

documents devoted to each component. Only some of them and

underlying intuitions are included in the following paragraphs.

1) Import Volume Function

The determinants usually adopted in the volume of imports are

domestic demand, a term in competitiveness formulated as the

relative price of domestic production compared to import prices

(usually calculated, excluding energy) and a term in productive capital

utilization. Usually, the cyclical economic pressures on production

capacity are described by integrating this equation into the utilization

rates of domestic production capacity relative to those of key

partners. This ratio helps to identify a possible supply constraint

which is subject to the national economy. The expected sign of its

elasticity with respect to imports is positive: when utilization rates

are higher in Tunisia than in its main partners, the increased domestic

demand is directed towards foreign producers, thereby increasing

the volume of imports. Lastly, some models enrich the analysis by

incorporating non-price competitiveness such as effort in research

and development (for example the integration of the age of capital).

2) Export Price Function

In fixing their prices, Tunisian producers are alleged to have a margin-

driven attitude towards foreign and domestic markets alike.

Nevertheless, to cope with foreign competition, they also take

account of foreign prices when setting export prices. Hence, there

is a trade-off between the margin-driven attitude (passing on the

total fluctuations in unit cost9 to export prices, so as to maintain a

constant profit margin), and a competitiveness-driven attitude

(passing on the total fluctuations in foreign prices to export prices

in a bid to maintain competitiveness). This trade-off translates into

a long-term target expressed as a weighted average of foreign prices

and domestic costs.

3) Import Price Function

Importers conduct a trade-off similar to that of exporters: in order

to maintain profit margins, they index their selling price on Tunisian

territory to their production costs, approximated by foreign import

prices. However, in order to maintain their competitiveness in

relation to domestic products, they also take into account domestic

production prices. Unlike foreign export prices, the foreign import

price is derived from simple weighting, given that competition takes

place only on Tunisian territory and therefore does not take third-

country markets into account.

4) VAR Modelling from the Cointegration Equation

As demonstrated in the final document of Component 2.1 based

on a cointegration equation of the form:

X iT(t) = a0 y(t) + a1 [piE(t) - py(t)] + f0 [piT(t) - piE(t) ] + G. f1 [piT(t) - piE(t)

]+ c + ê(t),10 ,


9 An approximation of unit costs may be made by incorporating domestic production prices.10 piE ; European competitor prices for commodity i,, py : GDP prices, piT : Tunisian price for commodity i, y(t) expressed demand.

28


It is possible to proceed with the estimation of a VAR model, in order

to conduct forecast exercises.

However, estimating such a model for each sector is a huge task,

and it is probably possible only after a selection of the most important

sectors for analysis (or considering only Tunisian exports to Europe

as a whole).

5) Non-linearities

The long-term structural equation for exports in Section 3.2 is based

on the assumption of constant elasticities. However, various forms

of non-linearities or structural changes may be considered.

- Temporal variation in income elasticity

For example, one can consider that income elasticity depends on

the European economy: in the early stages of the economic cycle

(for example, when unemployment rate u is higher than the natural

rate û), the income elasticity may be higher than in the low phases

of the cycle (when the unemployment rate is below the natural

rate).

In order to model this process, a formalization based on nonlinear

smooth transition models (Smooth Transition) may be proposed.

- Temporal variation in price elasticity

The price elasticity of foreign trade may depend on the absolute

difference between Tunisian export prices and competitor export

prices piT(t) - piE(t).

Indeed, when the price differential is small, i.e. when [piT(t-1) - piE(t-

1)]² is close to zero (or a given threshold k), the price elasticity of

Tunisian exports may be assumed to be relatively low, whereas when

the price differential is huge, i.e. when [piT(t-1) - piE(t-1)]² departs

significantly from zero (or a threshold k), Tunisian exports will be

heavily dependent on fluctuations in relative prices.

In order to model such phenomenon, the following formalization

may be proposed:

Suppose the transition function G (.), bound between 0 and 1,

wherein [piT(t-1) - piE(t-1)] stands for price differential, k the threshold

beyond which it is advantageous for consumers to change the

content of their consumption basket and h> 0 a parameter driving

the velocity of transition between regimes:

G([piT(t-1) - piE(t-1)] , h , k) =

It is established that when the price differential is very high

(positively or negatively) with respect to the threshold k (on the

brink as –h{[piT(t-1) - piE(t-1)]² - k} tends to infinity), function G

tends to 1, whereas when the price differential remains low (in the

sense that the distance between the price remains close to the

threshold k), function G tends to 0.

Hence, the proposed transition function makes it possible to model

a change in price elasticity based on the absolute difference in relative

prices.

Thus, the export equation can then be written as follows:

X iT(t) = a0 y(t) + a1 [piE(t) - py(t)] + f0 [piT(t) - piE(t) ] + G. f1 [piT(t) - piE(t)

]+ c + ê(t)

with: G =

Since the value of G depends on the absolute difference in relative

prices, the estimation of this equation will yield price elasticity

values between f1+f0 (to which it tends when the price differential

is high) and f0 (toward which it tends when the spread between

the prices is small).

The estimation of the model may be carried out by the non-linear

least squares method or maximum likelihood method to determine

the value of unknown parameters f0, f1, a0, a1, c, h and k.

29

6) Quantitative Rationing by Supply or Demand

It has been stated in the foregoing that the Tunisian sector export

equation:

XiT(t) = a0 y(t) + a1 [piE(t) - py(t)] + a2 [piT(t) - piE(t) ] + c

actually describes a Europe-driven demand equation. From this

perspective, it may be relevant to define such demand equation by

stating it:

(1) DmiT(t) = a0 y(t) a1 [piE(t) - py(t)] + a2 [piT(t) - piE(t)] + c

Similarly, the sector’s export supply (i) is conventionally

modelled as11 :

SXiT(t) = gS YiT(t) [PiTX(t) / PiTX(t)]sT

Where:

SXiT(t): Tunisian export volume supply of product (i)

YiT: Tunisian total production volume of product (i)

PiTD: price index of product (i) on the Tunisian domestic market (in

local currency)

PiTX: Tunisian export price index of product (i), in local currency.

gS : scale parametersT: verifying elasticity of processing sT >0

Let, in logarithms:

(2) SXiT(t) = yiT + sT [piTX(t) - piTX(t)] + c1

In the case of perfect price flexibility, the balance between supply

(7) and demand (6) will be achieved through an appropriate

adjustment of export prices12 . However, if it is assumed that there

is some export price rigidity, the quantity exported will stand at least

between supply (6) and demand (7).

An achievable estimation of a quantitative rationing model of this

type can be conducted through a CES function as follows:

(3)

wherein the supply and demand functions are defined by equations

6 and 7. Indeed, for large values of parameter ρ, the CES function

operates as operator Min.

The graph below illustrates the behaviour of the CES function with

respect to two time-varying variables S and D, where ρ = 100. It

can be observed that the CES function goes well with the minimum

of S and D.

Although the estimation of a CES function is not hitch-free, it may

be possible to econometrically estimate equation 8, in conjunction

with defining equations 6 and 7.


11 cf. Annabi and al., 2003.12 However, it is worth noting that PiTX differs from PiT , given that it does not factor in foreign exchange conversion, or customs duties or other costs borne by European importersof commodity i.

30


2.2.4. Bibliography

Annabi N., Cockburn J., Decaluwé B. (2003), Formes Fonctionnelles

et Paramétrisation dans les MCEG, CREFA, Université de Laval.

De Boeff, S. (2000), Modeling Equilibrium Relationships: Error

Correction Models with Strongly Autoregressive Data, Political

Analysis, Vol 9, 14-48.

Dickey, D.A., and Fuller, W.A. (1981), Likelihood Ratio Statistics for

Autoregressive Time Series with a Unit Root, Econometrica, Vol 49,

pp 1057-72.

Engle, R.F., and Granger, C.W.J. (1987), 'Cointegration and error

correction: representation, estimation and testing, Econometrica,

Vol 55, pp 251-276.

Hurlin, C. (2001), L’Econométrie des Données de Panel, Ecole

Doctorale Edocif, Séminaire Méthodologique.

Narayan P.K. (2004), Reformulating Critical Values for the Bounds

F-statistics Approach to Cointegration: An Application to the

Tourism Demand Model for Fiji. Discussion Papers No. 02/04

Monash University.

Pesaran, M.H., Shin, Y., and Smith, R.J. (2001), Bounds testing

approaches to the analysis of level relationships. Journal of Applied

Econometrics, Vol 16, pp 289-326.

Wong, K. N. (2008), Disaggregated export demand of Malaysia:

evidence from the electronics industry. Economics Bulletin, Vol. 6,

No. 6 pp. 1-14.

2.3 Analysis of the Demand for Tunisian Goods

The purpose of the analysis is to identify "promising commodities" for

Tunisia. The methodology followed comprises two major parts. In the

first part, potentially "promising" commodities for Tunisia are identified.

The typology is based on four main criteria: level of exports (global and

vis-à-vis the European Union) by Tunisia; level of revealed comparative

advantage (RCA); and variation in exports and revealed comparative

advantage. For variations (either of exports or RCA), the reference

period is 2003-2008, so as to better identify the dynamics involved.

With respect to levels, the calculation was done by taking the mean

between 2006 and 2008, with a view to eliminating business cycle

variations. From the COMTRADE database at the most disaggregated

level (6-digit), and using the four criteria, 30 industries were identified,

representing 25% of Tunisia's exports in 2008.

While the first part focuses instead on Tunisian supply, the second

part lays emphasis on demand. Consequently, for the 30 industries,

the changes in supply for each industry were analysed.

2.3.1. Analysis of Tunisia’s Supply in terms ofcomparative advantages

The industry list provides relevant information on comparative advantages

by sector. First, at level 2, there are 7 agri-food industries (HS01 -

HS23), 3 inorganic chemical industries (HS28; phosphates), and several

W "commodities" derived from iron and steel (HS72), and electrical

machinery (85). Almost all of these industries have a positive trade

balance and for most of them, the import level is very low. They are

mainly exporting industries, with very low intra-industry trade. Regarding

the level of exports, 4 industries account fort a significant share (12%)

of Tunisian exports: 150910, 280920, 310310, and 853690.

31


Table 12: Les produits potentiellement “porteurs” de la Tunisie

Product Product Name RCA Exports Trade Balance Exp. Share

030239 Tunas skipjack or stripe-bellied bonito... 0.96 58,382.56 58,381.12 0.30%

040630 Processed cheese, 0.81 29,775.37 28,811.48 0.15%

150910 Virgin 0.98 574,217.43 572,761.60 2.97%

150990 Other 0.93 46,095.95 46,053.40 0.24%

151000 Other oils..... 0.97 27,738.70 23,783.47 0.14%

151710 Margarine, excluding liquid margarine 0.87 38,352.88 38,352.31 0.20%

230690 Other 0.93 9,216.26 9,216.26 0.05%�

251010 Unground 0.94 147,511.96 147,511.96 0.76%

280920 Phosphoric acid and polyphosphoric acids 0.97 725,131.54 639,897.35 3.75%

283525 Phosphates: Calcium hydrogenorthophosphate 0.98 64,320.84 64,298.90 0.33%

283526 Phosphates:-- Other phosphates of calcium 0.97 90,545.08 90,523.38 0.47%

310310 Superphosphates 0.99 626,892.66 626,892.66 3.25%

520839 Dyed :-- Other fabrics 0.76 11,031.69 -40,415.97 0.06%

611249 Women's or girls' swimwear 0.99 35,060.86 33,570.27 0.18%

621010 Of fabrics of heading No. 56.02 or 56.03 0.97 158,879.24 146,244.53 0.82%

721030 Electrolytically plated or coated with zinc 0.52 22,776.00 19,676.60 0.12%

721049 Otherwise plated or coated with zinc 0.59 116,334.99 91,388.48 0.60%

721491 Other :-- Of rectangular cross-section 0.65 14,617.94 5,840.73 0.08%

740620 Powders of lamellar structure; flakes 0.97 20,300.78 20,244.92 0.11%

847190 Other 0.80 81,425.79 72,110.28 0.42%

851750 Other apparatus, for carrier-current line systems or for digit... 0.80 161,590.92 139,091.27 0.84%

852812 Reception apparatus for television 0.23 170,515.88 165,936.32 0.88%

853180 Other apparatus 0.93 68,192.05 63,597.62 0.35%

853690 Other apparatus 0.87 578,730.27 191,882.98 3.00%

854430 Ignition wiring sets and other wiring sets 0.70 172,225.15 -10,089.72 0.89%

854459 Other electric conductors 0.76 239,581.68 192,839.69 1.24%

854890 Other 0.87 89,169.01 69,144.95 0.46%

902830 Electricity meters 0.89 36,870.47 36,411.57 0.19%

961210 Ribbons 0.84 25,165.10 14,132.50 0.13%

961390 Parts 0.98 22,963.88 20,068.90 0.12%

Total 3,578,159.80 0.23

32


2.3.2. Global Demand Trends for Tunisian Expo

After identifying successful Tunisian exporting industries, the second

stage attempts to assess the demand trend and compare it with

Tunisian supply. Global demand trend for the 30 industries identified

is shown in the graph below:

Graph 5: World Imports from World 2003, 2008

The graph describes global imports for these products for two

years, 2003 and 2008, and the relative level of, and variation in,

the demand for each product. Five industries have a global

demand much higher than the rest: 721049, 852812, 853690,

854430, and 8545459. Industries for which demand has

increased the most are 251010, 230690, 310310, 854459, and

280920. For each of these five industries, global demand has

increased by more than 300%. During this period, the increase

in total global imports was a little over 110%. Twelve industries

had higher demand and two industries (030239, 851750)

experienced a decline in global demand (34% and 43%). See

also table below.

33


Tableau 13 : World Trade with world 2003, 2008

Reporter Product Product Name 2003 2008 2008 % Variation

0 030239Tunas (of the genus Tunnus) skipjack or stripe-bellied bonito (Euthynnus (Katsuwonus) pelamis),exc lunding livers and roes : -Other

556,830.991 366,877.877 -0,341

0 040630 Processed cheese, not grated or powdred 1,310,064.957 2,383,704.616 0.820

0 150910 Virgin 2,534,737.640 4,907,778.465 0.936

0 150990 Other 788,101.796 1,256,834.297 0.595

0 151000

Other oils and theirfractions, obtained solely fromolives, wether or not refined, but not chemiccalymodified, including blends of these oils or fractionswith oils or fractions of heading N°.15.0

85,354.027 230,863.345 1.705

0 151710 Margarine, excluding liquid margarine 742,651.639 1,624,874.252 1.188

0 230690 Other 65,728.005 303,258.366 3.614

0 251010 Unground 741,393.439 3,792,910.516 4.116

0 280920 Phosphoric acid and polyphosphoric adds 1,760,366.957 7,202,665.315 3.092

0 283525Phosphates: -calcium hydrogenorthophosphate(“dicalcium phosphate”)

264,012.468 633,375.252 1.399

0 283526 Phosphate:-other phosphates of calcium 297,946.680 1,059,238.644 2.555

0 310310 Superphosphates 627,845.727 2,700,043.471 3.300

0 520839 Dyed:-other fabrics 682,166.423 742,462.914 0.088

0 611249Women’s or girls’ swimwear:-of other textilematerials

63,634.259 64,613.937 0.015

0 621010 Of fabrics of heading N°. 56.02 or 56.03 951,616.144 1,488,329.897 0.564

0 721030 Electrolytically plated or coated with zinc 3,541,817.987 7,085,701.724 1.001

0 721049 Otherwise plated or coated with zinc:-other 9,291,254.117 22,466,215.247 1.418

0 721491Other:-of rectangular (other than square) cross-section

751,400.926 2,279,521.149 2.034

0 740620 Powders of lamellar structure; flakes 112,428.911 132,925.421 0.182

0 847190 Other 4,831,012.368 6,967,781.969 0.442

0 851750Other apparatus, for carrier-current line systemsor for digital line systems

17,359,352.886 9,886,454.734 -0.430

0 852812

Reception apparatus, for teleevision, whether ornot incorporating radio-broad cast receivers orsound or video recording or reproducing appa-ratus:-colour

26,404,762.434 78,694,420.177 1.980

0 853180 Other apparatus 2,038,031.989 2,282,609.436 0.120

0 853690 Other apparatus 16,462,291.070 31,350,854.230 0.904

0 854430Ingnition wiring sets and other wiring sets of akind used in vehicules, aircraft or ships

14,839,577.745 23,516,802.519 0.585

0 854459Other electric conductors, fora voltage exceeding80v but not exceeding 1,000 v:-other

5,282,229.260 22,222,484.002 3.207

0 854890 Other 2,761,272.151 3,242,141.436 0.174

0 902830 Electricity meters 822,344.712 1,658,544.995 1.017

0 961210 Ribbons 1,345,815.367 1,706,126.056 0.268

0 961390 Parts 101,217.088 142,548.323 0.408

34


Henceforth, the question now is to know which Tunisia’s potentially

competitive exporting countries are. To isolate the major competitors

in the 30 sectors, the ten largest exporters in the world were first

identified for 2008, and for each of the sectors, and the number of

times each country appears as the largest exporter or importer in

the selected industries was counted.

Table 14: 2008 – Number of times each country appears as key exporter or importer

Country Export Count Import Count

CHN 21 8

DEU 18 21

FRA 16 23

TUN 15 2

USA 15 21

ITA 13 16

ESP 12 18

NLD 12 14

BEL 11 12

GBR 10 20

JPN 10

MEX 9 8

TUR 9 1

ISR 6 1

KOR 6 5

CZE 5 1

MAR 5 1

POL 5 7

TWN 5 3

After identifying the "competitor" countries, it is interesting to assess

the level of similarity in the structure of exports between these countries

and Tunisia. The revealing indicator is Finger-Kreinin (FK). It allows a

comparison between the export structures of two countries. If the

structures are identical, FK is equal to "1", in case both countries have

totally different structures, which means there is no commodity exported

by both countries, FK is equal to "0". Table 3.15 provides the FK

between Tunisia and all countries identified in the table above. It

appears the level of similarity is highest with Morocco and Mexico

(0346 and 0334), followed by Turkey, and then some European

countries. The level of similarity is lowest with Israel, Japan, Korea,

and Taiwan. An indicator value of 0346 can be interpreted as a degree

of similarity of export structures of approximately 34.6%. In comparison,

the level of similarity between the U.S. and the EU is about 65%.

35


Table 15: FK Index for Tunisia and a Group of Competitor Countries

Country 2003 2004 2005 2006 2007 2008

BEL 0.160 0.137 0.157 0.173 0.177 0.185

CHI 0.205 0.187 0.199 0.206 0.211 0.210

CZE 0.153 0.154 0.178 0.176 0.184 0.191

FRA 0.165 0.161 0.183 0.194 0.203 0.204

DEU 0.152 0.147 0.167 0.177 0.180 0.181

ISR 0.080 0.078 0.081 0.087 0.099 0.110

ITA 0.205 0.199 0.222 0.232 0.232 0.236

JPN 0.092 0.092 0.105 0.117 0.125 0.132

KOR 0.126 0.104 0.117 0.131 0.134 0.141

MEX 0.245 0.261 0.320 0.293 0.333 0.334

MAR 0.380 0.384 0.376 0.353 0.342 0.346

NLD 0.164 0.141 0.157 0.166 0.167 0.182

POL 0.174 0.166 0.192 0.198 0.200 0.202

ESP 0.192 0.188 0.206 0.214 0.217 0.231

TUR 0.268 0.250 0.267 0.270 0.267 0.265

GBR 0.207 0.194 0.212 0.225 0.238 0.251

USA 0.144 0.140 0.155 0.171 0.167 0.179

TWN 0.122 0.113 0.125 0.145 0.143 0.149

In dynamics, the level of similarity with all countries (except Morocco

and Turkey) increased. FK with Morocco declined slightly over this

period, indicating a differentiation in specialization by both countries.

The increase in the index is most significant with Japan, Mexico,

and Israel. Hence, as regards the exports structure, it seems the

biggest competitors globally are some European countries (Czech

Republic, France, Italy, Spain), and Morocco but less so at the end

of period.

Unlike the FK index, the RECPI index takes into account the level

of exports. The table below gives the RECPI for Tunisia and the 18

countries, while considering their global exports. The higher the

number, the stronger the competitive pressure.

The key competitor countries are Mexico, United Kingdom, United

States, China and Nederland. The country whose competitive

pressure on Tunisia has increased the most is Korea.

36


Table 16: Indice RECPI de la Tunisie etd’un groupe de pays concurrents

Country 2003 2004 2005 2006 2007 2008

BEL 3.36 1.99 2.48 3.18 2.19 2.55

CHI 4.95 3.65 4.82 5.62 4.08 3.99

CZE 0.33 0.27 0.37 0.39 0.30 0.35

FRA 2.52 1.56 2.04 2.55 1.56 1.72

DEU 4.04 3.10 3.82 4.52 2.95 2.80

ISR 0.12 0.07 0.06 0.07 0.06 0.12

ITA 3.39 2.31 2.77 3.32 2.33 2.22

JPN 1.44 1.21 1.49 1.81 1.36 1.68

KOR 2.08 1.30 2.08 2.86 1.93 2.59

MEX 9.07 9.30 13.44 14.73 13.01 11.04

MAR 0.48 0.39 0.63 0.33 0.22 0.41

NLD 4.02 2.10 3.02 4.20 2.87 3.72

POL 0.47 0.35 0.46 0.55 0.40 0.44

ESP 1.75 1.48 1.52 1.95 1.20 1.37

TUR 1.16 0.88 0.90 0.97 0.72 0.75

GBR 8.83 7.79 10.26 11.31 9.52 8.71

USA 3.94 2.45 3.27 4.64 3.01 4.78

TWN 1.24 0.87 1.28 1.70 1.12 1.23

The table below reviews the level of similarity for the five largest

Tunisian exports sectors at 2-digit HS level. They account for over

60% of exports. For each of these sectors, the comparison

involves the level of similarity between Tunisia and the 18 countries

identified as key competitors. In the table, for each sector, the

five countries with the highest level of similarity are shown in red,

and the most significant in "bold". There are noticeable differences

among the sectors regarding the most significant competitors.

Morocco is one of the five most significant competitors in 4

sectors. Three countries, namely: Turkey, Belgium, and China are

in 3 areas. However, the variance is quite high across sectors. If

electrical appliances are taken into consideration, many countries

will have fairly identical levels of similarity. However, as regards

the inorganic chemicals sector, Morocco stands out as probably

the most significant competitor. Consequently, it is important to

consider the policies in these countries and the development of

their trade in key sectors when formulating subsequent relevant

policies.

37

2.4 Impacts of Opening up Tunisia’s Economyon the Production System and Analysis of theBusiness Adaptation Process

Productivity growth achieved by an economy as a whole may ensue

from two main sources:

- The internal dynamics of (or peculiar to) businesses; and

- The reallocation processes, among which the following

distinctions must be made:

- Reallocation among businesses within the same sector (usually

from the least efficient to the most efficient). This is intra-sector

reallocation;

- Reallocation of businesses across sectors (inter-sector

reallocation); and

- Reallocation through the entry and exit of firms (if in-coming

businesses are more efficient than out-going ones, then the net

impact on the production system is positive).

The objective of this part of the study is to assess the contribution

of these mechanisms to the dynamics of the Tunisian economy.

From a methodological standpoint, it was agreed that the productivity

decomposition method be applied. This method specifically identifies

the extent to which productivity growths are attributable to increased

productivity within businesses or to the phenomenon of reallocation.


Table 17: Tunisia’s Competitors by Sector

Mineral Fuels Electrical Machinery Apparel & Clothing FertilisersInorganicChemicals

27 85 62 31 28

CZE 0.217 0.350 0.587 0.005 0.105

TRU 0.212 0.322 0.521 0.517 0.143

GBR 0.712 0.247 0.444 0.155 0.023

USA 0.244 0.231 0.468 0.575 0.070

BEL 0.254 0.265 0.513 0.077 0.172

CHI 0.306 0.237 0.461 0.307 0.134

FRA 0.214 0.342 0.434 0.032 0.016

DEU 0.214 0.284 0.504 0.010 0.024

ISR 0.217 0.139 0.283 0.219 0.300

ITA 0.246 0.316 0.442 0.078 0.033

JPN 0.214 0.216 0.426 0.006 0.023

KOR 0.213 0.167 0.259 0.226 0.038

MEX 0.911 0.298 0.501 0.750 0.183

MAR 0.212 0.360 0.509 0.742 0.660

NLD 0.218 0.221 0.540 0.041 0.060

POL 0.215 0.358 0.508 0.025 0.112

ESP 0.214 0.330 0.446 0.092 0.117

TWN 0.213 0.145 0.294 0.001 0.030

38


2.4.1. Descriptive Analysis of Tunisia’s IndustrialBusiness Database

All data in the database are derived from annual surveys conducted

by the National Institute of Statistics of Tunisia and made available

to ITCEQ. The database contains data on Tunisian industrial

sectors from 1997 to 2007. It has information on production,

intermediate consumption, permanent employment, seasonal

employment, activity sectors, the region and capital structure. The

transition to constant prices was done by using the production

price indices, value-added price indices and 5-digit intermediate

consumption price indices provided by the INS. The business

performance indicator used is labour productivity, determined for

each company by calculating the ratio of value added at constant

prices to the entire workforce, which comprises both the

permanent and seasonal workers.

By retaining only the industrial sector, the "raw" initial database has

16,442 remarks, representing 4,464 businesses. Once cleaned

(detailed cleaning procedure in Box 1), the unbalanced panel

database includes 15 202 remarks and 4206 businesses.

Table 3.18 shows the breakdown of sample firms by sector. The

most prominently represented are apparel (29%) and agri-food

industries (13%). They alone account for 42% of firms in the sample.

The automotive sector, however, only accounts for 2% of the number

of firms, followed by the chemicals & pharmaceuticals, rubber &

plastics sectors (4% each).

Table 18: Number of Businesses by Sector

Sector Number of Businesses In percentage

1 Agri-food and Tobacco 561 13%

2 Textile 262 6%

3 Apparel 1236 29%

4 Footwear and Leather 250 6%

5 Timber, Paper and Publishing 283 7%

6 Chemistry and Pharmaceuticals 179 4%

7 Rubber and Plastics 159 4%

8 Non-metal Material 314 7%

9 Metal Materials 320 8%

10 Electrical Machinery, Machines and Appliances 329 8%

11 Automotive Industry and other Transportation Equipment 102 2%

12 Furniture 211 5%

The breakdown of 4,206 businesses by size was conducted based

on the criterion of average total employment of each business by

using the quantiles method. The resultant breakdown classifies

companies with a total number of employees not exceeding 23 under

the "small" group. In other words, the first third of the sample firms

have, on average, a number of employees not exceeding 23. In the

"medium" group (which corresponds to the second third of

businesses), businesses have a number of employees strictly greater

than 23 and not exceeding 77. "Large" group (the last third of sample)

firms have more than 77 employees.

39

Table 3.19 shows the breakdown of businesses by size and by

the number of years of presence in the database. The first row

of the table shows, for instance, that among the 1,469 firms

present in a single year, half of them (i.e. 732) fall under the

"small" category, about a third (or 455) belong to the " medium"

category and 19% (i.e. 282 companies) are considered "large".

Therefore, arrival and disappearance from the sample (which, it

should be recalled, are not necessarily new businesses or

cessation of activities) concern more of small firms. The

distribution by size of companies present for five years generally

corresponds to breakdown by quantile of 4206 businesses in

the sample. Approximately 80% of companies in the database

for 11 or 10 years fall under the "large" business category. On

average, companies with over 77 employees are present in the

sample during a number of years higher than the "medium" and

especially "small" categories.


Table 19: Number of Businesses by Size According to the Number of Years of Presence in the Database

Number of years ofpresence in the database

Number Percentage

Total Small Medium Large Small Medium Large

1 1469 732 455 282 50% 31% 19%

2 561 195 205 161 35% 37% 29%

3 469 144 172 153 31% 37% 33%

4 327 84 104 139 26% 32% 43%

5 322 114 106 102 35% 33% 32%

6 318 68 109 141 21% 34% 44%

7 205 27 78 100 13% 38% 49%

8 178 17 67 94 10% 38% 53%

9 140 10 46 84 7% 33% 60%

10 128 3 24 101 2% 19% 79%

11 89 1 15 73 1% 17% 82%

Table 3.20 dwells on the breakdown by size of businesses in the

database within the various sectors. For example, the second part

of the table shows that the breakdown in sector 10 (Electrical

Machinery) is closest to the distribution of firms across the database.

30% of companies are indeed small-sized, 36% are medium-sized

and 35% are large. However, the figures in bold or underscored in

gray highlight where the percentages substantially differ from those

that correspond to all the sectors. The highest fall in bold and the

lowest percentages are highlighted in gray. It is found that companies

in the "small" category are relatively more present in the timber, paper

& publishing sector (58%), Agri-food industries (57%), furniture

manufacturing (51%) and metal materials (45%) sectors. However,

they are less present in two sectors: apparel (10%) and footwear

and leather (23%). It is in this leather and footwear sector alone that

“medium-sized” companies are relatively the most active, with a 40%

share. Furthermore, it is only in one sector (Agri-food industries) that

these "medium-sized" businesses are relatively less prevalent (21%).

Lastly, companies considered "large" are more strongly represented

in apparel (57%), but relatively less present in timber, paper &

publishing (13%), metal materials (17%) agri-food industries (22%).

40


The table below shows the breakdown of firms by capital structure

and by size. Among the 4,206 businesses in the sample, 126 (3%)

have part of their capital held by the State and 1,243 (30%) have

part of their capital held by foreign investors. The firms concerned

fall mainly under the "large" category (58% for capital held by the

State and 65% for capital held by a foreign entity).

Table 3.22 shows the number of businesses by major region. It is

found that the vast majority of businesses in the sample are located

in the district of Tunis, the North East and Central East. Only 6%

of the 4,206 firms are located in the North West, 3% are located

in the South (East and West) and 2% of the sample in the Centre

West.

Table 20: Breakdown of Businesses by Sector and by Size

Number of years of presence in the database

Number Percentage


1. Agri-food and Tobacco 561 317 119 125 57% 21% 22%

2. Textile 262 96 103 63 37% 39% 24%

3. Apparel 1236 124 406 706 10% 33% 57%

4. Footwear and Leather 250 57 101 92 23% 40% 37%

5. Timber, Paper & Publishing 283 163 82 38 58% 29% 13%

6. Chem. and Pharmaceuticals 179 68 62 49 38% 35% 27%

7. Rubber and Plastics 159 61 62 36 38% 39% 23%

8. Non-metal Materials 413 124 109 81 39% 35% 26%

9. Metal Materials 320 145 121 54 45% 38% 17%

10. Electrical Machinery 329 98 117 114 30% 36% 35%

11. Automotive Industry 102 35 28 39 34% 27% 38%

12. Furniture 211 107 71 33 51% 34% 16%


Number of years of presence in the database

Number Percentage


Businesses that have at least part of their capital held by the State

126(i.e. 3% of 4206businesses)

24 29 73 19% 23% 58%

Businesses that have at least part of their capital held by foreign private investors

1243(i.e. 30% of 4206businesses)

107 334 802 9% 27% 65%

41

Table 23 shows unweighted average productivity by corporate

capital structure. It is found that businesses whose capital is held

in whole or in part by the State or foreign investors have, throughout

the period, an average productivity higher than all businesses.

However, no causal link may be inferred, given especially that, as

shown above, these firms fall mainly under the "large" category of

businesses. It is therefore not surprising to observe a higher average

productivity for both categories of businesses.


Table 22: Number of Businesses by Major Region

Regions Number of Businesses In percentage

1. District of Tunis and North East 1829 45%

2. North West 246 6%

3. Centre East 1731 43%

4. Centre West 95 2%

5. South East and West 122 3%

Total 4023* 100%

Corporate Capital Structure Unweighted Average Labour Productivity

Businesses that have at least part of their capital held by the State 9.22

Businesses that have at least part of their capital held by foreign private investors

8.71

* 183 businesses did not provide information on their location. Hence, the figures available fall short of the total number of businesses in the database, which stands at 4206.* This is the unweighted average, on all 11 years, expressed in log.

Table 24 provides the weighted average productivity per year

(expressed in log), which is also shown graphically (Graph 6). During

the past 11 years, the labour productivity of Tunisian businesses

in our sample rose sharply. Labour productivity (weighted average)

increased from 9.42 in 1997 to 9.67 in 2006 (which is a 25%

increase) and 9.91 in 2007 (i.e. 49% increase, still with respect to

1997). With respect to annual growth rates, productivity declined

only between 2002 and 2003 (5%), between 2003 and 2004 (1%)

and between 2004 and 2005 (1%). The strong productivity growth

registered between 2006 and 2007 (+24%) is quite surprising and

should be considered with caution. Indeed, the year 2007 is

characterized by a significant turnover of businesses in the sample.

Table 23: Average Labour Productivity by Capital Structure

42


As shown in Table 24, 30% of firms in 2007 were never previously

present in the database. It seems that these arrival and

disappearance of businesses have greatly contributed to such

increase in productivity between 2006 and 2007. Although INS

uses a number of procedures to ensure the representativeness of

the samples surveyed, caution must be exercised in interpreting

results when working on databases which are not from censuses.

Furthermore, to avoid distorting the interpretations, some graphs

will be presented: (i) covering the entire period (i.e. from 1997 to

2007) and (ii) leaving out the last year (i.e. from 1997 to 2006).

Graphs 3.4a show the weighted average labour productivity by

sector, first between 1997 and 2007, and then between 1997 and

2006.

If 2007 is disregarded, it will be discovered that labour productivity

fell in five sectors: textiles (sector 2), apparel (sector 3), chemicals

and pharmaceuticals (sector 6), rubber and plastics (sector 7) and

automotive (sector 11).

For the first 3 sectors (textiles (2) , apparel (3) and chemicals and

pharmaceuticals (6), the strong productivity growth registered

between 2006 and 2007 helps to avoid the foregoing cuts and to

end up in 2007 with productivity levels higher than in the beginning

of the period (i.e. 1997). In Sector 7 (rubber and plastics), labour

productivity also increased sharply between 2006 and 2007, but

not enough to exceed the productivity level of 1997. In addition, a

glance at the graphs over the period 1997-2007 reveals that only

YearWeighted average productivity (in log)

Annual growth rate of weightedaverage productivity

Productivity growth rate withrespect to 97

1997 9.42

1998 9.45 3% 3%

1999 9.49 4% 7%

2000 9.58 9% 16%

2001 9.61 3% 19%

2002 9.66 5% 24%

2003 9.61 -5% 19%

2004 9.60 -1% 18%

2005 9.59 -1% 17%

2006 9.67 8% 25%

2007 9.91 24% 49%

Table 24: Labour Productivity Trend for all Businesses in the Sample

43

one sector, the automotive sector (sector 11), experienced a

significant drop in its labour productivity.

In contrast, Labour productivity increased in seven sectors: food and

agricultural (sector 1), leather and footwear (sector 4), timber, paper and

publishing (sector 5), non-metal materials (sector 8), metal materials

(sector 9), electrical machinery (sector 10) and furniture (sector 12).

Among these sectors, labour productivity growth is particularly marked

in the food and agriculture sector (1), timber, paper and publishing (5),

non-metal materials (8), metal materials (9) and electrical machinery (10).

Graph 3.4b shows the labour productivity trend by corporate capital

structure between 1997 and 2007 and depicts, in solid line, the

category of businesses whose capital is wholly held by the State,

and in dotted line, the category of firms with at least part of their

capital held by foreign investors.

It is clear that the productivity of domestic firms rose more sharply

than that of companies with foreign capital. In 1997, the

productivity of domestic businesses stood at 9.32. It increased

to 9.72 in 2006 and to almost 10 in 2007. For companies with

foreign capital, it increased from 9.58 in 1997 to 9.68 in 2006

and 9.82 in 2007.

Between 1997 and 2003, the productivity of companies having

foreign capital is higher than that of domestic firms. From 2003,

the reverse is true: labour productivity of domestic firms becomes

higher than that of firms with foreign capital.

This mind-boggling outcome is interesting, and requires a more

detailed specific analysis. Indeed, it is generally expected that

businesses owned in part by foreign investors will experience more

substantial productivity growth.


44


Graphs 6: Weighted Average of Labour Productivity Trend by Sector between 1997 and 2007 (in log)Between 1997 and 2006 (in log)

Between 1997 and 2006 (en log)

45


Graphs 7: Weighted Average of Labour Productivity Trend by Size between 1997 and 2007 (in log)Between 1997 and 2006 (in log)

Between 1997 and 2006 (en log)

Graph 8: Weighted Average of Labour Productivity Trend by Corporate Capital Structure (in log)

46


2.4.2. Productivity Decomposition Analysis

2.4.2.1. Definitions and Methodology

Labour productivity growth in businessesmay be related to either:

- unanticipated cyclical changes in demand by businesses, which

generally account for frequent "unintentional" slumps in labour

productivity;

- labour market rigidities that can slow down the adaptation of the

number of employees (upward or downward) to changes in

production; or

- a set of firm-specific decisions that can lead to improved

productivity. These include, for example, improving the standard

of employee training, investing in the procurement of more efficient

machines, use of better quality inputs, corporate reorganization,

redundancy-related decisions, etc.

The phenomena of reallocation may ensue from inter-sector

changes (some sectors develop while others stagnate or decline),

or intra-sector changes, i.e. market share variations as well as

corporate arrivals and disappearances occur within each sector.

For a long while, because only sector data (from either domestic

or international sources provided by UNIDO) were available, the

reallocation analysis focused on inter-sector changes, given that

the homogeneous firm assumption posited by both traditional

international trade theories and the New Trade theory (Krugman,

1979, Helpman and Krugman, 1987) does not help to explain,

theoretically, the possibility of intra-industry reallocation. The recent

development of the "New New Trade" theory, initiated in particular

by Melitz (2003), and characterized by consideration of the

heterogeneity of businesses within sectors, justified theoretically,

that the analysis should focus on changes within firms. Access to

the individual databases of companies helped to develop empirical

analyses in furtherance of these theoretical advances. Hence, the

main lessons learned from recent theoretical and empirical

developments in the literature are only within the same industry.

There are companies that can be very outstanding, owing to their

size, degree of integration into the international economy, level of

productivity, etc. and that, in this context, any change (trade reform,

business environment, change in international demand, increased

competition, etc..), will impact differentially on these businesses

and necessarily engender reallocations within sectors. The

predominant concept from the literature is that these intra-industry

reallocations would be of much greater magnitude than those

occurring between sectors. In this theoretical framework, Melitz

(2003, cited above) has shown, for instance, that opening up to

international trade leads to increased market shares for businesses

that were initially the most productive to the detriment of the less

productive ones, which disappear or see their market share

dwindle. For these authors of "New New Trade", changes in the

aggregate productivity of an economy are due mainly to the

reallocation of such phenomena within industries, especially when

it comes to savings open to international trade.

In the literature, the 3 main methods used are those of Foster,

Haltiwanger Krizan (FHK, 1998 and 2001), Griliches and Regev

(GR, 1995) and, more recently, Pavcnik (2002). Although the FHK

method is the most comprehensive, it requires, as does also the

GR method, knowledge of the arrival and disappearance of

businesses. Given that data on Tunisian firms do not allow for the

identification of "real" arrivals and disappearances, the only method

applicable in this case is Pavcnik's.

This decomposition method was applied (i) across the entire sample,

(ii) by sector, (iii) by size and (iv) by corporate capital structure. In all

cases, the results are shown in terms of change from the start year,

i.e. 1997. In the four tables below, the second column indicates

changes in aggregate productivity with respect to 1997. The following

two columns correspond to variations of the first and second term

in the decomposition. As required by the decomposition equation,

the sum, online, of columns (3) and (4) corresponds to column (2).

2.4.2.2. Findings of Business Survey Processing

Table 25 shows the outcome of the entire sample. Column (2), which

indicates changes in aggregate labour productivity for all businesses,

is the last column of Table 12 already presented above. It is found

47

that much of the productivity growth is derived from the reallocation

effect. In 2006, the 25% aggregate labour productivity growth rate

were due to the 8% from productivity growth within businesses,

17% from the reallocation of resources from less efficient to the

most efficient firms. In other words, 67% of the variation in aggregate

productivity over 10 years (97-2006) is due to the increase in the

covariance term. In 2007, the same share stood at 72%. Although

this covariance term did not increase regularly throughout the period,

it is always positive (except only for the first two years), which shows

that reallocation plays in the right direction, i.e. the most productive

firms are developing and/or the least productive ones have decreasing

market shares.


(1) Year(2) Aggregate Productivity Growthtedaverage productivity (in log)

(3) Variation in unweighted Productivity(First term)

(4)Variation in Covariance (Second term)

1997 0.000 0.000 0.000

1998 0.023 0.055 -0.031

1999 0.071 0.072 -0.001

2000 0.153 -0.038 0.191

2001 0.183 0.043 0.140

2002 0.235 0.124 0.112

2003 0.183 -0.126 0.057

2004 0.172 -0.109 0.062%

2005 0.164 -0.079 0.085

2006 0.249 0.081 0.168

2007 0.486 0.138 0.348

Table 25: Decomposition of Aggregate Productivity Growth for the Entire Sample

While it is true that, for the entire sample, reallocation contributed

significantly to aggregate productivity growth, it is worth

underscoring also that such assertion still needs to be verified in

the sectors. In fact, it is only in two industries (footwear and

leather and metal materials), that changes in the covariance term

are always positive throughout the period and higher than corporate

productivity. However, in six sectors, (textiles, chemicals and

pharmaceuticals, rubber and plastics, non-metal materials, electrical

machinery and furniture), corporate productivity increased, while the

covariance term had a negative impact on the variation of aggregate

productivity. In the timber, paper and publishing sectors, the dominant

effect is labour productivity growth within businesses. In the food

and agricultural sector, productivity growth is also due, throughout

the period, to the productivity growth of businesses, except for the

last 2 years (2006 and 2007) during which contributions from the

covariance term were particularly significant. Lastly, in two sectors,

the apparel and automotive sectors, these two terms (corporate

productivity and covariance), played a negative role on the change

in aggregate labour productivity.

The impact of reallocation contributed significantly to the aggregate

labour productivity growth for the "average" and "large" categories

of businesses. With the exception of 1998, the variation of the

covariance term was indeed always positive for these two groups of

48


businesses. For the "small" category, that term varied positively only

for four years (2000, 2001, 2005 and 2007). The strong growth of

the covariance term in 2007 should be considered with caution, given,

as already underscored above, the crucial survey sample rotation

particularly relevant to "small" businesses. These results also show

that "medium"-sized businesses, for the most part, increased their

unweighted labour productivity. It would be interesting to understand

the factors that prompted them to improve their efficiency and the

means by which they achieved it.

The results of the decomposition of aggregate labour productivity

by corporate capital structure shows, in the sample, that the

reallocation effect tended to contribute to aggregate productivity

growth only for businesses that are entirely domestic. For firms

with part of their capital held by foreign investors, the variation in

the covariance term is positive only for 4 years (2000, 2001, 2002

and 2007). With regard specifically to entirely domestic businesses,

the 40% increase in aggregate labour productivity in 2006 can be

broken down as follows: 7% accounts for labour productivity

growth within businesses and 33% is derived from the reallocation

effect. In 2007, the 68% increase in aggregate productivity ensued

from productivity growth within businesses (14%) and the

reallocation effect (54%).

2.4.2.3. Conclusion

Under the project, labour productivity of Tunisian businesses in the

industrial sector was analysed between 1997 and 2007, from a

sample of individual firms from annual surveys. The key findings of

this analysis are as follows:

First, the aggregate labour productivity of Tunisian businesses rose

sharply. It increased by 25% between 1997 and 2006 (and by 49%

between 1997 and 2007, although the past year should be

considered with extreme caution, given that 30% of the sample

was renewed);

Second, at the sector level, aggregate labour productivity

increased in seven industries (Agri-food, leather and footwear,

timber, paper and publishing, non-metal materials, metal materials,

electrical machinery and furniture). However, if year 2007 is

disregarded, aggregate productivity fell in five sectors (textiles,

apparel, chemicals and pharmaceuticals, rubber, plastics and

automotive);

Third, while unweighted average productivity throughout the

period is higher for "large" firms than for "average" ones, aggregate

productivity grew faster for "medium"-sized businesses than for

"large" firms. From 2003 to 2006, the aggregate labour productivity

of "average" businesses exceeds that of "large ones;

Fourth, the aggregate labour productivity of domestic firms rose

more sharply than that of companies with at least part of their

capital held by foreign investors; and

Lastly, the decomposition results highlighted the role of resource

reallocation from less efficient to more efficient businesses in

boosting aggregate labour productivity across the entire sample.

The 25% productivity growth rate between 1997 and 2006 is

accounted for by 8% in labour productivity growth within

businesses and 17% from the reallocation effect. This is true

especially for domestic firms and "medium-" and "large-"sized

firms. However, at the sector level, this result is only true for 2

industries (footwear and leather, metal materials). Labour

productivity growth within businesses involved a larger number of

sectors (Agri-food, textiles, timber, paper and publishing, chemicals

and pharmaceuticals, rubber and plastics, non-metal materials,

electrical machinery and furniture).

49

Disney, R., J. Haskel and Y. Heden (2003), Restructuring and

productivity growth in UK manufacturing, Economic Journal,

Vol. 113, No. 489, pp. 666 – 694.

Foster, L., J.C. Haltiwanger and C.J. Krizan (1998), Aggregate

productivity growth: Lessons from microeconomic evidence, Working

Paper 6803 NBER.


productivity growth: Lessons from microeconomic evidence, in Edward

Dean, Michael Harper, and Charles Hulten (eds.), New Developments

in Productivity Analysis, Chicago: University of Chicago Press.

Griliches, Z. and H. Regev (1995), Firm productivity in Israeli industry

1979-1988, Journal of Econometrics, 65, pp. 175-203.

Hall, B. H. and J. Mairesse (1995), Exploring the relationship between

R&D and productivity in French manufacturing firms, Journal of

Econometrics, Elsevier, 65(1), pp. 263-293.

Helpman E. and P.R. Krugman (1987), Market Structure and

Foreign Trade: Increasing Returns, Imperfect Competition, and the

International Economy, MIT Press Books, The MIT Press.

Krugman, P. R. (1979), Increasing returns, monopolistic competition,

and international trade, Journal of International Economics, 9(4),

pp. 469-479.

Melitz, M. (2003), The Impact of Trade on Intra-Industry Reallocations

and Aggregate Industry Productivity, Econometrica, 71, pp. 1695–1725.

Pavcnik, N. (2002), Trade Liberalization, Exit, and Productivity

Improvements: Evidence From Chilean Plants, Review of Economic

Studies, 69, pp.245-276.


Bibliography

50

3. Training

The training of ITCEQ executives was an essential component

of the project. Two training sessions, the first in November 2010,

and the second in April and May 2011, allowed ITCEQ and DEFI

experts to develop constructive cooperation ties. These training

sessions in Aix-en-Provence and Tunis fostered the transfer of

knowledge in quantitative methods and discussed matters of interest

to ITCEQ. ITCEQ experts have course materials, STATA programmes

and access to DEFI databases. The non-exhaustive training reports

presented below outline the themes and methods addressed.

3.1 Example of Econometrics Training Sessionand Macroeconomic Modeling

29 November: Estimations using the Eviews software

Morning and Afternoon:

DEFI experts: Marcel Aloy, Eric Heyer

This day was devoted to the application of econometric techniques

on time series data presented the previous days.

Plan for the day

1. Recap of the main econometric tests on time series data

2. Introduction to the Eviews software

3. Estimations of various behavioural equations on French data

a. ECM in 2 stages

b. ECM in 1 stage

Tuesday 30 November: Presentation of various domestic models



Throughout this day, the structure and properties of the domestic

models existing in France were reviewed by comparing, most

particularly, the OFCE model (emod.fr) to that of the Ministry of the

Economy (Mésange).

Plan for the day

1. Accounting framework

2. Pattern of causalities

3. Transparency of causalities

4. Model size

5. Key behaviours

a. Consumption

b. Investment

c. Employment

d. Wage-price setting

e. External trade

6. Multiplier

a. Mechanism

b. Decomposition by major aggregate

c. Multiplier over time

d. Why does it vary?

e. Multiplier during the crisis.

7. Modeling Instruments

a. Error correction model

b. Conventional writing

c. The long run and cointegration

d. Dynamics

Wednesday 1 December: Structural properties of modelling &

estimations on Tunisian data

Morning:


Throughout this morning, the structural properties of modelling were


51


reviewed with special emphasis on the notions of potential growth

and structural unemployment.

Plan for the morning

1. Potential production

a. Structural assessment

b. Filter assessment

2. Structural unemployment

a. Structural assessment

b. Filter assessment

3. Review of various assessments and their impacts on economic

policy

Afternoon:


During this afternoon, Tunisian data were used to estimate the

employment- and price-value added equations for the various sectors.

3.2 Example of Training on Business DataProcessing

Tuesday, 23 November


DEFI expert: Gilbert Cette

Presentation of the analyses of potential production and methods

of calculating labour productivity: (see Slides of working sessions in

Appendix).

Wednesday, 24 November


DEFI expert: Marion Dovis

1. Presentation of methods for cleaning databases (see slides in

Appendix),

2. Presentation and explanation of major commands for the

management of databases on Stata (see slides in Appendix),

3. Work on Survey Data: Contrary to expectations, ITCEQ does

not have a comprehensive database bringing together all available

variables from the Survey data. In fact, data exists, solely on

Excel files separated by year and type of variables. However,

clean-up has to be performed on a single database. This

observation warranted that the creation of a complete index

of variables be added to the initially planned programme13.

4. Establishment of a single database containing all the available data

from surveys: These data are on Excel files separated by year and

by category of variables (results status, employment, liabilities,

assets, capital assets, identification, accounting values and other

book values and suite). The transition from these Excel files to a

full database on Stata would require the:

- harmonization of all Excel files to make them comparable;

- establishment of correspondences between different questionnaires

(questionnaire change from 98);

- addition of price indices; and

- merger of all these files so as to have a complete STATA package.

This study would normally require at least 3 or 4 working days. To carry

13 The initially scheduled work programme required that these data should be ready on a STATA database.

52

out the scheduled Training component within the same timeframe

and, in general, meet the set objectives of this mission, this work

was done largely at night and during the weekend by DEFI experts

(mainly M. Dovis). To avoid loss of time, Mr. Dovis provided ITCEQ

executives with an initial version of the database on Stata on Friday

afternoon, (so they can do the preparatory work of cleaning, and

get trained on Stata for the application of the session on the key

commands used, etc.).

Thursday 25 November

Morning:

DEFI expert: Patricia Augier

Presentation and explanation of the various productivity

decomposition measures used in scientific literature.

ffAfternoon:

DEFI experts: Patricia Augier, Marion Dovis

Harmonization of the various initial databases:

(i) Discussion on the correspondence between the 1997questionnaire

and the one used from 1998.

(i) Preparation of the programming file for the merger of various

databases.

Friday, 26 November

Morning and afternoon:


1. Continuation of the preparation of the programming file for the

merger of various databases.

2. Preparation of a preliminary version of the database for ITCEQ

executives so that they could start the initial descriptive analyses

in preparation for cleaning programmes.

3. Labelling the variables available in the database.

Thursday, 29 November

Morning:


Finalization and verification of the merger of databases with all the

variables.

Afternoon:


1. Verification of correspondences between the 1997 survey

questionnaire and the one used from 1998, particularly for the

profit and loss account and for employment. A file specifying

the transition between both questionnaires was prepared. For

cases not brought up for discussion, the calculation of

correlation coefficients was verified. The findings of these

correlation tests are specified on the relevant transition file (see

Appendix).

2. Preparation of a clean-up programme.

Tuesday, 30 November



1. Preparation of the clean-up programme.

2. Analysis of Inventory data to assess the possibilities of its use.

3. Discussion on the choice of the most appropriate productivity

decomposition method. It was agreed that an intuitive method

be chosen based on what is intended to be highlighted.

Furthermore, the issue of not identifying in-coming and out-going

businesses on the market should be considered when choosing

the decomposition method.


53


Wednesday, 1 December



1. Verification of the clean-up programme and detailed Explanation

of the various stages.

2. Discussion on possible changes to be introduced from the

baseline programme proposed and implemented by DEFI

experts.

3. Discussion on decomposition methodologies.

Thursday, 2 November



1. Discussion on decomposition methodologies.

(i) Choice of the method to be used

(ii) Choice of software to be used for programming.

2. Analysis of mission accomplishments with respect to expected

objectives and programming of the latter part of the project (see

details of these points in the following paragraph).

54

4. Conclusions

i) The ITCEQ/DEFI Technical Assistance Programme, whose

initially planned duration was 4 months, has actually been

implemented over 8 months, mainly because of the events

of January 2011 in Tunisia. The scope of the project was

very wide and some aspects have only been partially

addressed. However, in these few cases, the theoretical and

methodological elements were covered, thereby paving the

way for subsequent empirical validations.

ii) The training component met the set objectives mutually

agreed upon with ITCEQ experts during the start-up mission.

• ITCEQ experts were trained in econometrics and macro-

economic modelling. With respect to Quantitative techniques,

several sessions were devoted to processing software, especially

Stata, given that ITCEQ experts were more familiar with Eviews.

• ITCEQ experts were trained in theoretical macroeconomic

modelling. On this occasion, they were informed on

macroeconomic models used in major European forecasting

institutions.

• Lastly, ITCEQ experts were trained in components 2.2 and 3.

- ITCEQ executives were trained on methods of cleaning micro-

economic data used in the literature;

- TITCEQ executives received training on the management of

databases on Stata;

- ITCEQ executives have a merged and therefore complete

database, with all the variables;

- ITCEQ executives have the clean-up programme on Stata, and

the cleaned database with all the variables ready for immediate

use and/or which can be modified. This programme was

designed in such a manner as to enable ITCEQ executives to

subsequently modify it to their liking;

- ITCEQ executives were trained on the decomposition methods

used in the literature;

- The implementation of decomposition methods may be carried

out using various software such as Excel, Stata or Gauss. DEFI

experts and ITCEQ executives agreed that it was appropriate to

choose software such as Stata or Gauss which enables

programming.

iii) At the substantive level, the outputs obtained by ITCEQ and

DEFI experts are worthy of interest and open numerous avenues

for the development of Tunisia’s economic model and provide

responses to the challenges facing the Tunisian economy.

The modelling of micro manufacturing sectors based on approaches

to computable general equilibrium models has given rise to the

estimation of parameters, leading to the following conclusions:

- Substitution between capital and labour is high (elasticity close

to unity, Cobb-Douglas), in the food and agriculture sector as

well as in the Mechanical and Electrical Industries (sectors 2 and 4),

both of which are highly export-driven. Substitution between

both factors is lower (elasticity close to 0.7, CES) in the MCCV

and Chemicals sectors. In other sectors, there is complementarity

between factors of production;

Dynamic simulations show that the Tunisian industrial sectors

are characterized by a relatively high rigidity in the adjustment

of factors: in the wake of demand or actual cost shocks, it

generally takes three to four years for a significant adjustment

to occur in the quantity of labour and capital;

- The global productivity growth rate of factors is heterogeneous.

It is relatively low in IAA and textiles and high in IME and the

chemicals industry;


55


- In all the sectors studied, there is a high elasticity of substitution

between skilled and unskilled employment14 ranging from

3.3 for sector 5 to more than 6 for sector 4. This outcome is

significant, apparently robust and relatively unexpected: It implies

in particular that a 1% decrease in the relative wage of skilled

workers versus unskilled workers would lead (in the long run) to

an increase in the number of jobs for skilled workers higher than

that of unskilled workers by 3% to 6%;

Consequently, these initial findings allow for the validation of both

the theoretical and econometric methodologies implemented. Hence,

based on the work done in this study, sector estimates of prices

and foreign trade - which still have to be done - must be conducted

without major difficulties.

iv) The findings of the estimates of export functions by sector,

conducted on time series data, provide important lessons, as

they highlight huge differences among sectors, both with respect

to the long-term price elasticities of the various sectors (and

hence their level of substitutability on the European market), and

to the dynamic behaviour of Tunisian exports in the wake of

relative price shocks and demand shocks (see Appendix).

Henceforth, the estimation of an aggregate export equation, even

if it meets econometric quality criteria, cannot underpin forecasts

and the formulation of a coherent industrial development policy.

The data panel estimates helped to meaningfully supplement the

estimates made for each sector considered separately.


panel estimation allows for a rather robust estimation of the

'average' dynamic behaviour of sector exports. It follows that,

whatever the specifications used for the measurement of relative

prices:

- The elasticity of Tunisian exports to Europe in relation to expressed







substantially greater than 1 year;

- The price elasticity of Tunisian exports to Europe stands at -0.6

to -0.73 both in the short and long terms; and

- Lastly, the residual plots of the various models estimated suggest



the sector approach. Indeed, the time series data reveal that the

hydrocarbons sector has much higher price elasticity than other

sectors, while the IAA sector has relatively high short-term

demand elasticities compared to other export sectors

v) The project allowed ITCEQ experts to refine their knowledge of

Tunisia's specialization in international trade. The analysis of

Tunisia's competitor countries, through access to international

databases and the application of the outcomes of international

trade theory incorporated into the software "Swift Trade" was

strengthened. The dynamic positioning of Tunisia with respect

to global demand and the identification of competitor countries

are now available at the HS6 commodity classification level;

vi) Under the project, labour productivity of Tunisian businesses in

the industrial sector was analysed between 1997 and 2007, from

a sample of individual firms from annual surveys. The key findings

of this analysis are as follows:

14 It should be noted that the distinction between skilled/unskilled workers here is understood in terms of degree and not the type of business: in our estimates, skilled workerswere defined as higher education graduates, while the unskilled comprise other individuals. The terms used in our presentation ("skilled workers" and "unskilled workers") aretherefore deliberately simplistic and highly schematic.

56

First, the aggregate labour productivity of Tunisian businesses rose

sharply. It increased by 25% between 1997 and 2006 (and by 49%

between 1997 and 2007, although the past year should be considered

with extreme caution, given that 30% of the sample was renewed).

Second, at the sector level, aggregate labour productivity increased

in seven industries (food and agriculture, leather and footwear, timber,

paper and publishing, non-metal materials, metal materials, electrical

machinery and furniture). However, if year 2007 is disregarded,

aggregate productivity fell in five sectors (textiles, apparel, chemicals

and pharmaceuticals, rubber, plastics and automotive).

Third, while unweighted average productivity throughout the period

is higher for "large" firms than for "average" ones, aggregate


"large" firms. From 2003 until 2006, the aggregate labour productivity

of "average" businesses exceeds that of “large” ones.


more sharply than those of companies with at least part of their

capital held by foreign investors.





accounted for by 8% in labour productivity growth within

businesses and 17% from the reallocation effect. This is true

especially for domestic firms and "medium-" and "large-"sized

firms. However, at the sector level, this result is only true for two

industries (footwear and leather, and metal materials). Labour

productivity growth within businesses involved a larger number

of sectors (agri-food, textiles, timber, paper and publishing,

chemicals and pharmaceuticals, rubber and plastics, non-metal

materials, electrical machinery and furniture).


Chapter I

Sector Modeling of Manufacturing Industries

in Tunisia

58


Table of Contents

59 Introduction

61 1. Theoretical Foundations

62 1.1 The ‘unit price of value added’: definition

62 1.2 Demand for Factors

63 1.3 Demand foe Skilled and Unskilled Labour

63 1.4 Determining the Price of Value Added

64 1.5 Apportioning Supply between Exports and Supply of Goods on the Domestic Market

65 1.6 Apportioning Demand between Imports and Domestic Goods

65 1.7 Determining the Price and Volume of Exports

66 1.8 Defining Equations

66 1.9 Defining Equations

68 2. Equations f the Model

68 2.1 List of Variables and Parameters of the Model

69 2.2 Log-linear Equations (long-term relations)

70 2.3 Equations in Levels

72 3. Econometric Estimations72 3.1 List of Variables and Parameters of the Model72 3.1.1 Employment, Capital and Price of Value Added74 3.1.2 Skilled and Unskilled Employment

75 3.2 Data 21

75 3.3 Estimation Method

75 3.4 Key Findings77 3.4.1 Employment, Response to a Demand and Cost Shock79 3.4.2 Capital Response to a Demand and Cost Shock

81 Conclusion

82 Bibliography

59


Introduction

In this paper, the modelling of a sector i producing a compositegood i is considered. This composite good can be either sold on

the domestic market or exported. Similarly, end users can purchase

this good on the domestic market from a sector i or import it.

The theoretical structure of the model hinges on principles widely

used in computable general equilibrium models (see Lofgren et al.,

2002):

Sector i is subject to the constraint of a two-factor production function

(labour L and capital K) of type CES, whose unit costs (respectively

W and Ck) are given at the sector level.

Production (QX = Y/iva, where iva represents value added in volume

per unit produced) from sector i is distributed between domestic

market sale (QD) at price Pd and export sale (QE) at price Pe. The

optimal distribution of production (QX) depends on the relative

price (Pd/Pe).

The proposed model quite naturally helps to consider as special

cases sectors whose production is intended solely for exports, or

alternatively, sectors whose production is entirely destined for the

domestic market.

Intermediate consumption is a constant fraction (ci) of the quantity

produced.

With respect to end users, the domestic absorption of goods typically

produced by sector i (QQ) is broken down between domestic sector

production (QD) and similar import products (QM). The absorption

distribution between both components depends on the relative price

of domestic products (Pd) with respect to the price of imported

products (Pm).

As concerns the conventional specifications of computable general

equilibrium (CGE) models, the peculiarities of this model are as follows:

- First, it is considered that the sector is subject to a demand

constraint on the goods market , so that optimal demand for

production factors (K and L), is realized under the constraint of

a given demand (Y / iva), which is determined by other sources

such as the level of domestic absorption (QQ),. Therefore, it is

essentially a demand model and not a supply model;

- Second, a distinction is introduced, within the labour factor (L),

skilled labour (Lnq), unskilled labour (LNQ),. Each of these two

types of labour characterized by a specific productivity index ,

as well as an own wage rate (Wq and Wnq respectively), assumed

to be exogenous in the sector; and

- Third, a foreign export demand is introduced, which determines

the export equilibrium price (Pe), whereas it is generally assumed

to be exogenous in the CGE models.

Section I presents the long-term linearized equations, whose

theoretical underpinnings are explained in Section II.

Indeed, the CGE models have the advantage of offering high internal

consistency, thus involving long-term solutions that may be

considered theoretically satisfactory. In return, these models generally

do not provide a description of the dynamic adjustment process

towards long-term solutions.

From this viewpoint, it is suggested in Section III that the estimation

of model equations be conducted within the broader dynamic

framework of error-correction equations.

For purposes of illustration, consider the equation for determining

60

the price of value added pyit (in logarithms):

pyit = π0 (ckit ― gki.t) + (1― π0) (wit ― qli.t) + py015 (23’)

where:

gki;: autonomous capital productivity growth rate;

qli: autonomous labour productivity growth rate.

The econometric estimation may be conducted in 2 stages:

1. Estimation of long-term relation:

pyit = π0 (ckit ― gki.t) + (1― π0) (wit ― qli.t) + py0 + Zt

2. Estimation of error-correction model:

∆pyit = a0 + a1∆pyit-1 +a2∆ckit +a3∆wit ― bZt-1 + εt

Where: ∆xt: variation of xt (the latter being expressed in logarithm).

Naturally, the lag order of explanatory variables will be determined

by conventional econometric criteria.

This type of specification is used to display short-term elasticities

(a2, a3) subsequently different from long-term elasticities (π0 and1― π0 in our example).

In addition, it helps to measure the adjustment speed towards long-run

equilibrium (which is related to parameter b), the latter being conditioned

by the various bottlenecks affecting the functioning of markets.

The advantage of the error-correction model is that it guarantees

convergence towards long-run equilibrium (provided that the empirical

results confirm the validity of equation 23 in our example), -

symmetrically - The Granger representation theorem states that a

long-term relationship (as defined in the cointegration theory),

necessarily admits representation in the form of ECM.

Following this methodology, which is particularly preferred in the

macroeconomic model of the Bank of England16, it is therefore possible

to propose and estimate a sector modelling that is both theoretically

coherent and consistent with observable dynamic processes.


15 Lower case letters designate variables in logarithm16 This involves implementing the usual unit root tests, e.g. Dickey and Fuller (1981).

61


1. Theoretical Foundations

Assume that each sector i produces a composite good i. The

model seeks to determine the long-term equilibrium quantities

and prices, based on the principles used in computable general

equilibrium models. In the theoretical model, the sector index (i), will

be omitted to simplify the notations.

1.1 The 'unit price of value added’: definition

Value added in value is decomposed according to:

Pyt Yt = Wt Lt + Ckt Kt (1)

where:

Py: price of value added of the sector

Y: value added in volume of the sector

W: average wage rate of the sector

L: employment of the sector

Ck: capital cost of the sector

K: capital stock of the sector

and the sector decomposition of production in value is written as

follows:

Pxt QXt = Pyt Yt + Pcit CIt + TAXt ― SUBVt (2)

where:

QX: production in volume of sector i

Px: production price of the sector

Pci: unit price of intermediate consumption of the sector

CI: quantity of intermediate consumption

TAX: taxes paid

SUBV: subventions received

Let:

CIt = ci.QXt

where ci is the intermediate consumption coefficient

SUBVt = sv.QXt (3)

where sv is the amount of subventions received per unit produced

TAXt = tva.pyt.Yt (4)

where tva is the value added tax rate

Yt = iva.QXt

where iva is the value added in volume per unit produced

The ‘price of value added’ of sector (Py) is defined below by the

expression:

Pyt = [Pxt ― Pcit ci + sv] / [iva (1+tva)] 5)

Given that the price of value added (Py) is subsequently determined

based on marginal costs, this expression helps to define the sector’s

unit production price:

62

Pxt = Pyt iva (1+tva) + ci Pcit ― sv (6)

Intermediate consumption of sector (i) may be disaggregated

depending on their sector of origin (j), which implies:

CIt = QXt SOMME(ci[j]) (7)

and consequently:

ci = SOMME(ci[j]) (8)

Pcit = SOMME(Pqt [j] ci[j]) / ci (9)

1.2 Demand for Factors

The technical constraint of production is a CES standardized

production function (cf. particularly Klump et al., 2008):

(10)

where:

Y0 , L0 , K0 : value added in volume, employment and capital stock

at reference date t=0

Alt , Akt : labour and capital productivity indices

π0 = Ck0 K0 / Py0 Y0: share of capital remuneration in value addedat reference date

Py0 Y0 = Ck0 K0 + W0 Y0: value added at reference date

First-order conditions for the optimum lead to the following relation:

Kt / Lt =( Akt /Alt ) σ − 1 (Ckt / Wt) − σ (K0 / L0) 1 − σ [π0/(1- π0)] σ

Capital Stock and Investment

If businesses are subject to a demand constraint (Y given), the stock

of capital deduced from the production function and optimum

condition will be:

(11’)

The log-linear approximation of this equation may be expressed

as follows:

kt = yt ― σ (ckt ― wt) + (σ― 1) gk.t + k0 (11’)

where Ckt is the cost of capital utilization based on Hall-

Jorgenson:

Ckt = Pkt (it+1 ― pk’t+1 + δ) (12)

with:

Pk: investment price,

i: nominal interest rate,

δ: capital depreciation rate,

pk’: capital price rate of change.

Given the capital stock evolution equation:

Kt = It + (1―δ).Kt-1 (13)

Optimal investment will be:

It = Kt ― (1―δ)Kt-1 (14)

It will suppose that independent capital productivity follows a

deterministic trend:

Akt = (1+gk)t (15)

Demand for Labour

Market-constrained demand for labour is written as follows:

(16)


63


Which is expressed in logarithms as:

lt = yt ― σ (wt ― pyt) + (σ ― 1)gl.t + l0 (16’)

It will suppose that independent labour productivity follows a

deterministic trend:

Alt =(1+gl)t (17)

1.3 Demand for Skilled and Unskilled Labour

Total employment is supposed to be a CES combination of skilled

labour (Lq) and unskilled labour (Lnq):

(18)

where:

Alqt, Alnqt: skilled and unskilled labour productivity index

γ0 = Wnq0 Lnq0 / W0 L0: share of unskilled labour remuneration in

total remuneration at reference date

W0 L0 = Wnq0 Lnq0 + Wq0 Lq0: wage bill at reference date

At the optimum, demands for skilled and unskilled labour are

respectively written as follows:

Lnqt = Lt (Alnqt / Alt)κ−1 (Wnqt /Wt ) −κ (Wnq0 /W0 ) κ−1 γ0 (19)

Lqt = Lt (Alqt/ Alt) κ−1 (Wqt/Wt)−κ(Wq0/W0) κ−1 (1-γ0) (20)

where:

Wnq: unskilled wage rate

Wq: skilled wage rate

It is supposed that the independent labour productivity of each type

follows a deterministic trend:

Alnqt =(1+glnq)t (21)

Alqt =(1+glq)t (21a)

Hence, the log-linear equations of demands for unskilled and skilled

labour are respectively written as follows:

lnqt = lt ― κ (wnqt ― wt ) + (κ ―1)( glnq ― gl).t + lnq0 (19’)

where glnq.t is a trend term that measures the independent

productivity of unskilled labour.

lqt = lt ― κ (wqt ― wt ) + (κ ― 1)( glq ― gl).t + lq0 (20’)

where glq.t is a trend term that measures the independent productivity

of skilled labour.

The average wage rate of the sector will be a combination of skilled

and unskilled wage rates:

(22)

From where the following log-linear equation is deduced:

wt = (1―π0)(wqt ― glq.t) + π0 (wnqt ― glnq.t) + gl.t + w0 (22’)

1.4 Determining the Price of Value Added

Given the accounting relation (equation 1):

Pyt Yt = Wt Lt + Ckt Kt

where Y is production of the sector, W is the average wage bill of

the sector, L is the total employment level of the sector (equation

16), Ck is the cost of capital utilization (given in equation 12) and K

is the capital stock (equation 11).

The price of value added is deduced from the optimal demand of

factors:

64

(23)

which leads to the following log-linear approximation:

pyt = π0 (ckt ― qk.t) +(1― π0) (wt ― ql.t) + py0 (23’)

1.5 Apportioning Supply between Exports andSupply of Goods on the Domestic Market

Domestic production in value of sector i is decomposed in exports

in value and domestic production in value:

Px t (Y t / iva) = Pet QEt + Pdt QDt (24)

where:

Px: total production price of sector i

Y / iva: total production of sector i

Pe: export price of sector i

QE: exports supply of sector i

Pd: price of goods of sector i on the domestic market

QD: supply of goods on the domestic market of sector i

If domestic production:

- Is not sold on the domestic market: Y/iva = QE

- Is not sold for export: Y/iva = QD

The processing function of domestic production between the two

destinations is written as follows (cf Annabi et al., 2003):

(25)

Where:

Y0 / iva,QD0 , QE0 : total production of sector i, supply of goods on

the domestic market of sector i and exports supply of sector i at

reference date t=0

μ0 = Pe0 QE0 / (Py0 Y0 / iva): share of exports in value in the production

of sector i at reference date

Px0 Y0 / iva = Pd0 QD0 + Pe0 QE0: production in value at reference

date

-∞ < π < 0: processing elasticity

At the optimum, the maximization of revenue (24), constrained by

the processing function (25), helps to determine the supply of exports

and the supply of goods on the domestic market:

QEt = (Yt / iva) (Pet / Pxt) ―π(Pe0 / Px0) π ― 1 μ0 (26)

QDt = (Yt / iva) (Pdt / Pxt) ―π(Pd0 / Px0) π ― 1 (1-μ0) (27)

Let:

qet = yt + π (pxt ― pet) + qe0 (26’)

qdt = yt + π (pxt ― pdt) + qd0 (27’)

with π<0

and production price is deduced from the optimal apportioning of

supply:

Given that production price (Px) is determined based on the price

of value added (equation 6) and that the export price is determined

by balancing supply and export demand (equation 37), the expression

of the price of good i on the domestic market may be deduced from

this equation:

which leads to the following logarithmic approximation:

pdt = [ 1/ (1― μ0)]pxt ― [μ0/ (1― μ0)] pet + pd0 (28’)


65


1.6 Apportioning Demand between Imports andDomestic Goods

Domestic absorption in value for good i is decomposed in imports

in value and domestic production in value:

Pqt (1―tq) QQt = Pmt QMt + Pdt QDt (29)

where QQ is domestic absorption (sum of public and private

consumption, investments, intermediate consumption), tq is tax

rates, Pm is the price of import goods of sector i and QM is the

volume of good imports of sector i.

The Armington aggregation function of absorption between both

components is written as follows (cf Annabi et al., 2003):

where:

QQ0, QD0 , QM0 : total absorption of goods i, demand for goods i

on the domestic market and demand for imports i at reference

date t=0

λ0 = Pm00 QM0 / [Pq0 (1―tq0) QQ0]: share of imports in value in the

absorption at reference date

Pq0 (1―tq0) QQ0 = Pd0 QD0 + Pm0 QM0: absorption in value at

reference date

∞ > α > 0: substitution parameter

If good i:

- Is not produced on the domestic market: QQ = QM

- S not imported : QQ = QD

The minimization of total cost (30), constrained by the Armington

function, implies that at optimum, the demand for imports and for

goods i on the domestic market will respectively be:

QMt = QQt (Pmt / Pqt (1―tq)] ―α [Pm0 / Pq0 (1―tq0)]α ―1 𝜆0 (31)

QDt = QQt [Pdt / Pqt (1―tq)] ―α [Pd0 / Pq0 (1―tq0)]α ―1 (1― 𝜆0) (32)

or, in the form of logarithm:

qmt = qqt ― α (pmt ― pqt + tq) +qm0 (31’)

qdt = qqt ― α (pdt ― pqt + tq) + qd1 (32’)

And the absorption price is deduced from the optimal apportionment

of absorption:

(33’)

which implies the following log-linearized equation:

pqt = tq + α0(pmt ― pdt) + pdt + pq0 (33’)

1.7 Determining the Price of Quantities traded onthe Domestic Market

The price and quantities traded on the domestic market are deduced

from equations 27 and 32.

The quantities of good i traded on the domestic market as a result

of balancing supply and demand are given in the following equation:

QDt = (Yt /iva) α /(α ―π) QQt―π /(α ―π) (Pxt / Pqt (1-tq)] α π /(α ―π) C1 (34)

where: C1 = [(1― π0) ―π /(α―π) (1―μ0) α /(α―π) Px0 α (π ― 1) /(α ― π) [Pq0

(1―tq0)] ( 1― α)[ ―π /(α―π)] ] / Pd0

which gives the log-linear equation:

qdt = [α /(α ― π)] yt ― [π /(α ― π)]qqt + [α π /(α ― π)] (pxt ―

pqt + tq) + d0 (34’)

The equilibrium price of good i on the domestic market resulting

from the equalization of supply and demand is as follows:

66


Pdt = (iva QQt / Yt ) 1/(α ― π) [Pqt (1-tq)]α /(α ―π) Pxt ―π /(α ―π) C0

where:

C0 = Pd0 Px0 (π ―1) /(α ―π) [Pq0 (1―tq0)]( 1― α)/(α ― π) [(1― π0)/(1―μ0)][1/(α ― π)

Since the definition of domestic price is given in equation (28), the

volume of value added (Y) can be determined to meet the equilibrium

of good i on the domestic market:

(35)

where: C2 = [(1―π0)/(1―μ0)] [Pq0(1―tq0)]1―α Px0π―1

i.e., by linearizing:

yt = qqt + α(pqt - tq) ― π pxt + (π ― α)pdt + y0 (35’)

1.8 Determining the Price and Volume of Exports

Assuming that foreign demand (QE) for goods of sector i is expressed

as follows:

QEt = cx Y*t [(1― te) EXR P*t / Pet] π (36)

where Y*t is foreign income and P*t is the foreign price index

(expressed in foreign exchange units, FCU).

The ratio Pe / [(1 ― te) EXR] helps to convert the export price into

foreign currency units, and takes into account current import taxes

on export markets (te).

which can be expressed in logarithms as follows:

qet = cx0 + y*t + π (p*t + exrt ― te ― pet) (36’)

The export price is deduced from equations 26 and 36 which helps

to balance supply with demand for exports, as well the equilibrium

export volume:

Pet = Pxtπ / (π ―π) [P*t EXR (1― te)] ―π / (π ― π) [(μ0 Yt) / (cx iva Y*t)]1/(π―π)

(Pe0/Px0) (π ― 1) / (π ―π) (37)

QEt = [Pxt / (P*t(1-te)EXR )] ―π π / (π―π) (cx Y*t) π / (π―π) (μ0 Yt / iva ) ―π / (π―π)

(Pe0/Px0)π (1 ― π) / (π ―π) (38)

i.e., in logarithms:

pet = [ π / (π ― π)]pxt ― [ π / (π ― π)] (p*t + exrt ― te) + [1/

(π ―π)] (yt ― y*t) + pe0 (37’)

qet = ― [π π / (π ― π)] [pxt ―(p*t + exrt ― te)] +[ π /(π ―π)]

y*t ― [ π /(π ―π)] yt + qe0 (38’)

1.9 Defining Equations

Import Prices

Pmt = Pwmt (1+tm).EXRt (39)

Pm: import price of good i LCU

Pwm: import price of good i FCU

tm: import tax rate of good i

EXR: exchange rate LCU to FCU

i.e., in logarithms:

pmt = pwmt + tm + exrt (39’)

Export price in foreign currency (FCU)

Pwet = Pet / (1―te).EXRt (40)

Pe: export price of good i in local currency (LCU)

Pwe: f.o.b. export price of good i in foreign currency (FCU)

te: export taxes of good i (levied by importing countries)

EXR: exchange rate LCU to FCU

i.e., in logarithms, the expression of export price in local currency:

pet = pwet ― te + exrt (40’)

67


Domestic demand for composite good i:

QQti = Cti + SOMME(Ijti) + GOVti (41)

where:

Ci: final consumption of goods i

Iji: investment of sectors j in goods i

GOVi: public expenditure in good i

68

2. Model Equations

2.1 List of model variables and parameters

Structural parameters are presented in Greek characters or

in lower-case, unsubscripted letters. The same applies for

economic policy parameters.

Endogenous Variables

Py: price of value added of the sector

Y: value added in volume of the sector

W: average wage rate of the sector

L: total labour of the sector

Lnq: unskilled employment of the sector

Lq: skilled employment of the sector

K: capital stock of the sector

QX: production in volume of the sector

Px: production price of the sector

CI: quantity of intermediate consumption used in the sector

TAX: taxes paid in the sector

SUBV: subventions received by the sector

Px: total production price of the sector

Pe: export price of goods i

QE: exports in volume of the sector

Pd: price of good of sector i on the domestic market

QD: sales in volume of goods on the domestic market of the sector

Pm: import price of good i in local currency (LCU)

Pe: export price of good i in local currency (LCU)

Pwe: f.o.b. export price of good i in foreign currency (FCU)

QM: imports in volume of goods i

Exogenous Variables (in the sector)

Wnq: unskilled wage rate

Wq: skilled wage rate

Pci: unit price of intermediate consumption in sector i

QQ: domestic absorption in volume (sum of public and private

consumption,

investments, intermediate consumption) in goods i

Y*t: foreign income

P*t: foreign price index (in foreign currency).

Pwm: import price of good i in foreign currency (FCU)

EXR: exchange rate (local currency LCU to foreign currency FCU)

te: export tax rates of good i (levied by importing countries)

Alt , Akt : labour and capital productivity index

Alqt , Alnqt : skilled and unskilled labour productivity index

Ckt: capital cost of the sector, decomposed as follows:

Pkt: investment price,

it: nominal interest rate,

pkt’: capital price rate of change.

Economic Policy Parameters

tq: production tax rate of good i

tm: import tax rate of good i

sv: subventions received per unit produced

tva: value added tax rate


69




∞ > σ > 0: substitution elasticity between capital and labour

(production function)

∞ > κ > 0: substitution elasticity between unskilled employment and skilled employment (aggregation function)

-∞ < π < 0: processing elasticity (processing function)

∞ > α > 0: substitution parameter (Armington function)

-∞ < - π < 0: foreign demand price elasticity

ci: intermediate consumption coefficient

iva: value added per unit of volume produced

2.2 Log-linear Equations (long-term relations)

The level variables are represented in uppercase letters and variables

in logarithms in lower case.

Latin letters indexed by a 0 (e.g. x0) represent the constant terms

of long-term regression equations, while the Greek letters represent

the structural parameters of the theoretical model.

The numbering of the equations refers to those of the theoretical

model presented in Chapter II.

Capital stock of the sector:

kit = yit ― σ (ckit ― pyit) + (σ−1) gki.t + k0 (11’)

where: Ckit is the cost of capital utilization based on Hall-

Jorgenson:

Ckit = Pkit (it+1 ― pk’t+1 + δ) (12)

with:

Pk: investment price,

i: nominal interest rate,


pk’: capital price rate of change.

and gki measures the rate of change of independent capital

productivity.

Total employment of the sector:

lit = yit + σ (wit ― pyit) + (σ ― 1)gli.t + l0 (16’)

where gli measures the rate of change of independent labour

productivity.

Unskilled employment of the sector:

lnqit = lit ― κ (wnqit ― wit ) + (κ ―1)( glnqi ― gli).t + lnq0 (19’)

where:

glnq measures the rate of change of independent unskilled labour

productivity.

wnq is the unskilled wage rate

Skilled employment of the sector:

lqit = lit ― κ (wqit ― wit ) + (κ ― 1)( glqi ― gli).t + lq0 (20’)

where:

glq measures the rate of change of independent skilled labour

productivity.

wq is the skilled wage rate

Average wage rate of the sector:

wit = (1―π0)(wqit ― glqi.t) + π0(wnqit ― glnqi.t) + gli.t + w0 (22’)

Price of value added of the sector:

pyit = π0 (ckit ― gki.t) + (1― π0) (wit ― gli.t) + py0 (23’)

Price of goods sold on the domestic market:

70

pdit = [ 1/ (1― μ0)]pxit ― [μ0 / (1― μ0)] peit + pd0 (28’)

Absorption price of good i:

pqit = tqi + π0 (pmit ― pdit) + pdit + pq0 (33’)

Import volume of good i:

qmit = qqit ― α (pmi

t ― pqit + tqi) + qm0 (31’)

where qq is the domestic absorption of good i

Sales volume of sector i on the domestic market:

qdit = [α /(α ― π)] yit ― [π /(α ― π)]qqit + [α π /(α ― π)] (pxit ―

pqit + tqi) + d0 (34’)

Value added of the sector:

yit = qqit + α(pqit - tqi) ― π px it + (π― α)pd it + y0 (35’)

Export sales price of sector i:

peit = [ π /(π ― π)] pxit ― [ π /(π ― π)] (p*t + exrt ― te)+ [1/(π ―π)]

(yit ― y*t) + pe0 (37’)

where:

p* : foreign price index

y* : real foreign income index

Export volume of sector i:

qeit = ― [π π /(π ― π)] (pxit ― p*t ― exrt + te) + [ π / (π ―π)] y*t

― [ π / (π ―π)] yit + qe0 (38’)

Defining Equations

Import prices of goods i:

pmit = pwmi

t + tmi + exrt (39’)

Export prices of goods of sector i in foreign currency (LCU):

pweit = peit + tei - exrt (40’)

2.3 Level Equations

Intermediate consumption of sector i:

CIit = cii QXit

where ci is the intermediate consumption coefficient

Subventions received by sector i:

SUBVit = svi QXit (3)

where sv is the amount of subventions received per unit

produced

Taxes paid by sector i:

TAXit = tvai pyit Yit (4)

where tvai is the value added tax rate

Total production volume of sector i:

QXit = Yit / ivai

where iva is the value-added volume per unit produced

Production price of sector i:

Pxit = Pyit ivai (1+tvai) + cii Pciit ― svi (6)

Intermediate consumption of sector (i) disaggregated by sector of

origin (j):


71


CIit = QXit SOMME(cii [j]) (7)

And consequently:

cii = SOMME(cii [j]) (8)

Pciit = SOMME(Pqit [j] cii [j]) / cii (9)

Investment of sector i:

Iit = Kit ― (1―δ)Kit-1 (14)

Domestic demand for composite good i:

QQti = SOMME(CIjit) + Cti + SOMME(Ijti) + GOVti (41)

where:

CIji: intermediate consumption by sectors j of goods i

Ci: final consumption of goods i

Iji: investment by sectors j on goods i

GOVi: public expenditure on good i

72

3. Econometric Estimations


Capital stock of sector i:

kit = yit ― σ (ckit―pyit) + (σ―1) gki.t + k0 (11’)

Total employment in sector i:

lit = yit +σ (wit―pyit) + (σ―1)gli.t + l0 (16’)

Price of value added of sector i:

pyit = π0 (ckit―gkI.t) + (1―π0) (wit―gli.t) + py0 (23’)

where:

kit volume of capital stock of sector i (logarithm)yit volume of value added of sector i (logarithm)(ckit � pyit) cost of capital utilization of sector i in real terms (logarithm)gki annual capital productivity growth rate of sector i lit number of hours worked in sector i (logarithm)(wi

t � pyit) actual wage rate in sector i (logarithm)gli annual labour productivity growth rate in sector ipyit value added price index of sector i (logarithm)

Table 1: Estimated factor demand equations (long term)

As is mentioned in Table 1, the econometric estimation is performed

on the logarithm of the various model variables (capital, employment,

wages, cost of capital, price of value added).

Regarding the calculation of the cost of using capital (ckit), two

measures were tested:

- The first is to calculate the cost of using capital as defined in

Hall-Jorgenson (i.e. by calculating the sum of real interest rate

and capital depreciation rate); and

- The second uses the unit cost of using capital, derived from the

accounting allocation of value added.

Based on equations derived from the theoretical model, this

section presents the estimation of the production block of

the Tunisian manufacturing industry decomposed into six sectors.

The estimated equations are firstly those regarding total

employment, capital stock, price of value added, and secondly,

those pertaining to employment, broken down into skilled and

unskilled workers.

III.1 Estimated equations

3.1.1 Employment, capital and price of value added.

We conducted the estimation of factor demand functions arising from

the CES production function (equation 10). This amounts to estimating

the demand for capital equations (equation 11'), labour (equation

16') and the price of value added (equation 23’) (Table III.1).

73


Among these two possible specifications, the second proved to be

much more relevant and it was therefore chosen for the purpose of

the estimates.

Second, it is worth noting that employment (lit) can be measured in

two ways:

- Either in terms of stock, i.e. by retaining the number of people

employed;

- Or, in terms of flows, by retaining the total number of hours

worked (hours per person multiplied by the number of people

employed).

To best measure the impact of a wage change, the second of these

measures was selected: indeed, following a change in the relative

cost of labour, firms may be led to modify the number of employees,

or the average number of hours worked.

Lastly, it is worth noting that each production factor is supposed

to assume the growth rate of its own productivity (gki measures the

rate of change of independent capital productivity and gli measures

the rate of change of independent labour productivity in each sector

i). In accordance with standard practice, it is assumed that growth

rate is constant over time, which means that in the long run, the

'technical progress' incorporated into factors follow a deterministic

trend.

Regarding the estimation of such growth rates of technical progress

(respectively denoted gl and gk), three types were tested: the first

is that of a neutral technical progress as defined by Harrod, i.e. it

only increases work efficiency (gl > 0 and gk = 0). The second is

that of technical progress which only increases the efficiency of

capital (neutrality as defined by Solow, gl = 0 and gk > 0). A final

form induces an increase common to both factors (Hicks-neutral,

gl = gk > 0).

As presented in Table III.1, the estimated equations represent long-

term relationships, provided the usual conditions for cointegration

are met (the variables of the models must be integrated to order 1

and residuals of equations integrated to order 0).

From this point of view, one of the objectives of the estimate is to

assess the long-term sensitivity of capital stock trend and

employment levels to changes in real factor costs (wages and cost

of capital), through the estimation of elasticities of substitution (σ)

between capital and labour, in each sector under review. Thus, a

high elasticity of substitution will lead to a substantial increase in the

capital/labour ratio as a result of an increase in the relative price of

factors (salary/cost of capital).

To take into account the dynamics of the model in the shorter term,

i.e. outside the long-run equilibrium, the additional parameters of the

model were estimated under an error-correction model- ECM (Table 2).

Capital stock of sector i:Notation: Δxt = xt ― xt-1

Capital stock of sector i:Δkit = a0 + a1Δ yit + a2Δ yit-1 + a3Δ(ckit ― pyit) + a4Δ(ckit-1 ― pyit-1) + a5Δ kit-1

― γk [ kit ― yit + σ (ckit ― pyit) ― (σ ― 1) gki.t ― k0] + εkt

Total Employment in sector i:Δli

t = b0 + b1Δ yit + b2Δ yit-1 + b3Δ(wit ― pyit) + b4Δ(wi

t-1 ― pyit-1) + b5Δ lit-1― γl [ lit ― yit ― σ (wi

t ― pyit) ― (σ ―1) gli.t ― l0] + εlt

Price of value added of sector i:Δpyit = c0 + c1Δ ckit + c2Δ ckit-1 + c3Δwi

t + c4Δwit-1 + c5Δpylit-1

― γp [pyit ― π0 (ckit ― gki.t) ― (1― π0) (wit ― gli.t) ― py0] + εpt

Table 2: Factor Demand Equations estimated (ECM)

74

The equations presented in Table 2 represent the form being

tested in all generality. To improve the specification, parameters not

significantly different from zero were not retained in the final model,

and some dummy variables were introduced.

In its current form, the ECM model is a significant asset for the

dynamics: short-term elasticities (coefficients a1, a2, …. , a5 ; b1, b2,

…. , b5 ; c1, c2, …. , c5) may be introduced, and differ from long-

term elasticities (σ, π0 ,…).

Similarly, the error-correction model helps to measure buoyancy (γk, γl,

γp) toward long-run equilibrium, i.e. the proportion of a deviation in the

long run may be absorbed in a given period: given that the elementary

period here is one year, it is expected that these coefficients are close

to unity if the dependent variable is not characterized by strong viscosity.

3.1.2. Skilled and unskilled employment

In addition, the demand for skilled/unskilled labour was estimated

(equations 19 and 20 of the theoretical model).

The objective here is to assess the short- and long-term impact of

a change in the relative wage on the structure of (skilled/unskilled)

employment.

At the theoretical level, companies benefit from using relatively

more skilled workers - in spite of higher wages - on condition that

they have productive characteristics distinct from those of unskilled

workers (or, in other words, that skilled and unskilled workers are

not close substitutes). In the opposite case, companies will benefit

from using cheaper labour.

As a result, the key parameter for measuring the impact of a change

in the relative wage on the employment structure will be the elasticity

of substitution between the two categories of employees.

Table 3 shows the two long-term employment demand equations

(only one which is estimated because of redundancy), and this

for each of the 6 manufacturing sectors: the elasticity of

substitution between skilled and unskilled workers κ is noted

therein.


Unskilled employment in sector i:lnqit = lit ― κ (wnqit ― wi

t ) + (κ ―1)( glnqi ― gli).t + lnq0 (19’)

Skilled employment in sector i:lqit = lit ― κ (wqit ― wi

t ) + (κ ― 1)( glqi ― gli).t + lq0 (20’)

where:

glnq rate of change of independent unskilled labour productivitywnq unskilled wage rate glq rate of change of independent skilled labour productivitywq skilled wage rate

Table 3: Employment Demand Equations by Qualification (long term)

It should be noted that the distinction between skilled/unskilled

workers here is understood in terms of degree and not the type

of business: in our estimates, skilled workers were defined as higher

education graduates, while unskilled workers comprise other

individuals. The terms used in our presentation ("skilled workers"

and "unskilled workers") are therefore deliberately simplistic and

highly schematic.

To complete the estimation, long-term equations were once more

placed in the broader context of error correction models. The generic

form of the dynamic equations estimated is given in Table III.4. As

above, the non-significant short-term coefficients were removed from

the final equations.

75


3.2 Data

The estimation period spans from 1983 to 2009. Data were collected

from the annual national accounts for the Tunisian economy. Six

manufacturing sectors are analysed, namely:

Sector 2: Agri-food Industry (IAA);

Sector 3: Ceramic Building Materials and Glass Industries (MCCV);

Sector 4: Mechanical and Electrical Industries (EMI);

Sector 5: Chemical Industries (CHEMISTRY);

Sector 6: Textile Apparel and Leather Industries (THC); and

Sector 7: Miscellaneous Industries (MISCELLANEOUS).

Before conducting the econometric estimation, the order of

integration of the series used was tested using the Augmented

Dickey-Fuller (ADF) test. The results show that the entire series tested

are all integrated to order 1.

3.3 Estimation Method

The low number of observations available (at most 27) and the

presence of common parameters to be estimated in the various

long-term equations (elasticities of substitution, the growth rate of

technological progress) prompted the simultaneous estimation of

these equations.

Consequently, the following two-stage process was adopted:

3. In stage one, long-term relationships were estimated, per level,

for equations 11’, 16’, 20’ and 23’ through a simultaneous

equations system using the SUR (Seemingly Unrelated

Regression) method.

4. In stage two, 4 ECMs (Error-Correction Models) were estimated

by imposing therein relations estimated in stage one as long-

term solution.

3.4 Key Outputs

Before commenting on the outputs, it is worth noting that all ECMs

have satisfactory statistical properties. LM tests result in the rejection

of the hypothesis of autocorrelation in the residuals of these

equations. These residuals are homoskedastic under the White test

and the ARCH test. The functional form of the equation passed the

Reset test. Lastly, according to the Jarque Bera test, the residuals

of all equations are normally distributed.

The key outputs are summarized in Table III.5:

• There would be a relatively high substitutability between capital

and labour in 4 sectors as follows: the elasticity of substitution

is close to unity in the Agri-food Industry as well as in the

Mechanical and Electrical Industry (sectors 2 and 4). The elasticity

of substitution is close to 0.7 in the Ceramic Building Materials

and Glass and Chemicals industries (sectors 3 and 5). However,

it is worth noting that the value of these elasticities is probably

Notation: Δxt = xt ― xt-1

Unskilled employment in sector i:Δlnqit = d0 + d1Δ lit + d2Δ lit-1 + d3Δ(wnqit ― wi

t) + d4Δ(wnqit-1 ― wit-1) + d5Δ lnqit-1

― γnq [ lnqit ― lit + κ (wnqit ― wit ) ― (κ ― 1)( glnqi ― gli).t ― lnq0 ] + εnqt

Skilled employment in sector i:Δlqit = g0 + g1Δ lit + g2Δ lit-1 + g3Δ(wqit ― wi

t) + g4Δ(wqit-1 ― wit-1) + g5Δ lqit-1

― γq [lqit ― lit + κ (wqit ― wit ) ― (κ ― 1)( glqi ― gli).t ― lq0] + εqt

Table 4: Employment Demand Equations by Qualification (ECM)

76

over-estimated by including working hours in labour. In other

words, this does not involve elasticity between capital and labour,

which is being estimated here, but between capital stock and

work hours.

• However, there seems to be a strong complementarity between

capital and labour in the last two sectors under review, namely

Textile, Apparel and Leather and miscellaneous industries (sectors

6 and 7). This implies that in these sectors, and unlike the previous

ones, a variation in the relative cost of labour relative to capital

will have little long-term impact on capital intensity (capital stock

per worker).

• As per our estimates, it seems that the formulation of a Hicks-

neutral technical progress is the only one to be accepted in the

6 sectors studied. In sectors 2, 6 and 7, the estimated growth

rate of technical progress (gk and gl parameters, which are

identical in the Hicks-neutral technology parameters) is between

1 and 2% per year. It is close to 2.7% in sector 4 and above

5% in sectors 3 and 5.

• In all the sectors studied, there is a high elasticity of substitution

between skilled and unskilled employment ranging from 3.3 for






workers by 3% to 6%. In other words, a tighter wage gap

between skilled and unskilled workers would be, according to

this estimate, an effective means of improving the "employability"

of graduates. One possible interpretation of this result is that,

on average, graduates would not have enough specific expertise

that would suitably distinguish them from unskilled workers, given

that both categories are consequently considered by businesses

as substitutes rather than complements in the production

process.

• In conclusion, it should be recalled that long-term elasticities

with respect to quantities were fixed to unity, for obvious reasons

of theoretical consistency: thus, when the total amount of work

increases by 1%, the amount of hours put in by skilled and

unskilled workers increases identically by 1% (all things being

equal). Similarly, when demand for goods (measured by value

added in volume) increases by 1%, the amount of capital and

labour increases identically by 1%.


Table 5: Main Outcomes for Sector Estimates

Sector 2 3 4 5 6 7


Elasticity of substitution K/L 0.99 0.74 0.98 0.67 0.06 0.14

Elasticity of substitution Skilled/Unskilled labour 5.40 4.10 6.01 3.29 4.67 5.55

Hicks-Neutral technical progress…

… (%) .1.97 5.11 2.68 5.28 1.06 1.67

ECM Parameters

Employment buoyancy 0.40 0.21 0.32 0.12 0.21 0.31

Capital buoyancy 0.05 0.13 0.07 0.06 0.14 0.04


77


To illustrate the dynamic process of the different variables and their

sensitivity to economic determinants, the assessment of sector

responses to some economic shocks was conducted.

3.4.1. Employment Response to Demand and CostShock

Two shocks were simulated:

1. The first is a 1% increase in value added

2. The second shock is a 1% increase in real wages

The outcomes of these simulations are summarized in Table 6 and

in the graphs below.

Table 6: Dynamic Response of a 1% VA and Real Wages (RW) Shock on Employment

T 1 year 2 years 3 years 4 years Long term

IAA VA 0.31 0.46 0.62 0.71 0.79 1

RW -1.16 -0.63 -0.97 -0.84 -0.94 -0.98

MCCVVA 0.85 0.88 0.38 0.52 0.62 1

RW 0 -0.77 -0.76 -0.75 0.74 -0.73

IME VA 0.44 0.62 0.74 0.82 0.88 1

RW -0.21 -0.46 -0.62 -0.73 -0.81 -0.97

CHEMI VA 0.19 0.30 0.28 0.31 0.32 1

RW -0.22 -0.06 -0.01 -0.20 -0.28 -0.67

THCVA 0.37 0.50 0.61 0.69 0.76 1

RW -0.47 -0.38 -0.31 -0.26 -0.21 -0.05

MISCEL VA 0.53 0.68 0.78 0.85 0.90 1

RW 0 -0.04 -0.07 -0.09 -0.11 -0.13


78


Graph 1: Employment variation (in %) following a 1% increase in value added

Graph 2: Employment variation (in %) following a 1% increase in real wages

3.4.2. Capital Response to Demand and Cost Shock

Two shocks were also simulated for capital:

3. A 1% increase in value added

4. A 1% increase in the real cost of capital.

The results of these simulations are summarized in Table 7 and in

the graphs below.

79


Table 7: Dynamic Response of a 1% VA and actual cost (CKR) shock on Capital Stock.

T 1 year 2 years 3 years 4 years Long term

IAA

VA 0.318 0.51 0.71 0.83 0.91 1

RW -0.10 -0.27 -0.40 -0.50 -0.64 -0.98

MCCV

VA 0.17 0.36 0.58 0.78 0.93 1

RW -0.14 -0.29 -0.45 -0.61 -0.73 -0.73

IME

VA 0 0.18 0.42 0.65 0.83 1

RW 0 -0.07 -0.21 -0.39 -0.56 -0.97

CHEMI

VA 0 0.15 0.25 0.39 0.52 1

RW -0.03 -0.15 -0.24 -0.34 -0.43 -0.67

THC

VA 0.27 0.50 0.67 0.79 0.87 1

RW -0.008 -0.028 -0.036 -0.042 -0.047 -0.05

MISCEL

VA 0 0.04 0.10 0.17 0.24 1

RW -0.04 -0.07 -0.08 -0.09 -0.10 -0.13


80


Graph 3: Capital Stock Variation (in %) following a 1% increase in value added

Graph 4: Employment variation (in %) following a 1 % increase in actual cost

81


This study lays the groundwork for a macro-sector modelling

of the Tunisian economy. It was delimited to the theoretical

specification of manufacturing industries and the econometric

estimation of the production block (demand for factors and price

of value added).

At the theoretical level, the various equations are based on the

principles used in computable general equilibrium models (CGEM),

while at the econometric level, estimates utilize the properties of

error-correction models (ECM).

Accordingly, the proposed model meets the requirements for the

internal consistency of long-term solutions and external coherence

with respect to observable data. The estimates conducted present

a certain number of outcomes worthy of interest:

- For most industries, the elasticities of substitution are close to

unity, which coincides with the traditional assumption of Cobb-

Douglas production functions. For some sectors, on the contrary,

(Textiles, Apparel, Leather and Various Industries), it is rather the

assumption of a production function with complementary factors

that seems to be appropriate.

- Overall, skilled and unskilled labour are revealed to be close

substitutes, thereby making the proportion of both types of input

very sensitive to changes in their relative cost.

- Dynamic simulations show that the Tunisian industrial sectors

are characterized by a relatively high rigidity in the adjustment

of factors: in the wake of demand or actual cost shocks, it

generally takes three to four years for a significant adjustment

to occur in the quantity of labour and capital.

Consequently, these preliminary findings make it possible to validate

both the theoretical assumptions and econometric methodologies

implemented. Hence, based on research work conducted under

this study, the estimation of sector prices and foreign trade - that

still have to be done - must be finalized in the relatively short run.

Conclusion

82

Bibliography

Annabi N., Cockburn J., Decaluwé B. (2003), Formes Fonctionnelles

et Paramétrisation dans les MCEG, CREFA, Université de

Laval.

Harrison R., Nikolov K., Quinn M., Ramsay G., Scott A. and

Thomas R. (2005), The Bank of England Quarterly Model,

www.bankofengland.co.uk/publications/beqm/

Klump R., McAdam P., Willman A. (2008), Unwrapping some euro

area growth puzzles: Factor substitution, productivity and

unemployment. Journal of Macroeconomics 30, 645–666

Lofgren H., Harris R.L., Robinson S. (2002), A standard computable

general equilibrium model in GAMS. Microcomputer in Policy

Research 5, International Food Policy Research Institute.


Chapter II

Modeling Tunisia’s Exports by Sector

84


Table of Contents

85 Introduction

87 1. Presentation of Europe-Bround Sector Export Equations87 1.1 European Demand for Tunisia Imports: General Framework87 1.1.1 European Demand for Imports of Commodity87 1.1.2 Distribution of European Imports of Commodity88 1.1.3 Equation for Tunisian Exports of Commodity (i) to Europe88 1.2 Estimating the Equation for Tunisia Sector Exports: Error-Correction Model88 1.2.1 Error-Correction Model89 1.2.2 Estimation of dynamic Data Panel Models

90 2. Econometric Estimation of Tunisian Sector Export Equations90 2.1 Data91 2.2 Estimatio Method91 2.3 Key Outputs91 2.3.1 Agriculture94 2.3.2 IAA sector97 2.3.3 Building Materials, Ceramics and Glass Sector (MCCV)99 2.3.4 Mechinacal and Electrical Industries Sector101 2.3.5. Chemicals Sector104 2.3.6 Textile Apparel and Leather Sector106 2.3.7 Miscellaneous Sector «Miscel»108 2.3.8 Hydrocarbons and Refined Products Sector110 2.3.9 Total Goods112 2.3.10 Summary

114 3. Econometric Estimation of Tunisian Sector Export Equations in Dynamic Data Panel114 3.1 Data117 3.2 Estimation Method117 3.3 key Outputs117 3.3.1 ADL Model in Levels in tne Case of EU Prices (LPUE)118 3.3.2 ADL Model in Levels in the Case of Competitor Country Prices (LPCON)120 3.3.3 ADL Model in First Differences in the Case of EU Prices (DLPUE) 121 3.3.4 Error-Correction Model (ECM) in the Case of EU Prices (LPUE)123 3.3.5 Error-Correction Model (ECM) in the Case of Competitor Country Prices (LPCON)124 3.3.6 Error-Correction Model (ECM) in the case of Competitor Prices (LPCON)

127 4. Possible Extensions of Econometric Analysis127 4.1 A Foreign Trade Model ( in volume and in price)127 4.1.1 Volume Import Function127 4.1.2 Export Price Function128 4.1.3 Import Price Function128 4.2 VAR Modeling from the Cointegration Equation128 4.3 Non-linearities128 4.3.1 Income elasticity Variation Over time129 4.3.2 Price Elasticity Variation Over time130 4.4 Quantitative Rationing by Supply or by Demand

132 Conclusions133 Bibliography

85


Introduction

This study sets out to analyse the determinants of Tunisian

exports, broken down into 8 sectors. Accordingly, it helps to

quantify the dynamics of sector exports as a result of variations

in expressed demand and relative prices.

This application will be delimited to the study of Tunisian exports

towards Europe, which represents the bulk of Tunisian exports,

based on annual data for the period 1988-2008.

The first section proposes a model of European import demand

for Tunisian commodities which, according to conventional

import functions17, leads to the introduction of 3 explanatory

variables:




- Tunisia’s competitor price index on the European market.

Hence, this theoretical model is an extension of conventional

export models18 .

An econometric estimation methodology for this model is proposed

based on error-correction models. This technique helps to distinguish

between short-term elasticities and long-term elasticities of European

sector imports.

The second section presents the results obtained by the application

of this econometric methodology on Tunisian exports of goods,

broken down into eight sectors (plus total exports of goods). By

constraining the long-run elasticity of demand to a unit value, this

application is intended to reflect the changing market shares of

Tunisian exports to European countries.

Thereafter, the third section gives some estimates through dynamic

data panel methods.

Lastly, in the fourth section, some interesting modelling extensions

are discussed, which may be undertaken as a continuation of

previous work:

- First, estimated equations will be supplemented by a model

of sector prices and quantities (imports and exports), which

may help to determine nominal flows and trade balances

by sector;

- Second, some modelling and estimation avenues are proposed

for the purpose of introducing nonlinearities. These are intended

to take into account the possibility of time-varying elasticity. For

example, price elasticities may be lower when price differentials

are small and more significant when price differentials increase,

due to transaction costs; and

- Lastly, supply constraints will be integrated into the model. Indeed,

the previous modelling is limited to the specification and

17 See for example N. Annabi, Cockburn J., B. Decaluwé (2003) for an overview of the microeconomic foundations of the demand functions.18 See for example Wong (2008) for a recent application to the case of Malaysia.

86

estimation of European sector demand for Tunisian commodities.

However, especially in the presence of price rigidities, exports

may be limited by available supply and, consequently, actual

exports may not always correspond to European demand for

them. In this regard, a method has been proposed for the

estimation of a quantitative rationing model.


87

1. Presentation of Europe-Bound Sector Export Equations

1.1 European Demand for Tunisian Imports:General Framework

1.1.1. European demand for imports of product (i)

The European demand function for the importation of commodity

(or sector) i is as follows:

(2) MiE = θΕ Yσ1 (PiE / Py)σ2

Where:

MiE: European import volume of commodity (i),

Y: European overall income volume

PiE: European import price index of commodity (i)

Py: European overall production price index

θΕ is a scale parameter

In principle, elasticities verify the following conditions:

Income elasticity: σ1 >0

European import price elasticity: σ2 <0

At the theoretical level, the following points may be underscored:

- In maximizing a utility function based on a CES function, income

asticity (σ1) is necessarily equal to1 (cf. Annabi et al., 2003).

- The price elasticity of demand for imports (σ2) is related to the

elasticity of substitution between European imports and domestic

production. If imports to Europe are perfect substitutes for local

production, price elasticity tends to minus infinity, whereas, if

they are complementary, price elasticity will tend towards

zero.

Additional remark:

- This equation may be used to analyse the imports of a European

country considered separately, rather than those of the entire

zone. In this case, Y and Py will respectively represent the income

and domestic price index of the European country considered.

- Other demand indicators such as consumption or investment

may be considered, rather than overall GDP (Y), if it is considered

that some commodities exported to Europe only meet consumer

demand (or alternatively investment demand). Consequently, the

consumer or investment price index should be retained as overall

price index, depending on the appropriate demand indicator.

1.1.2. Distribution of European imports of product (i)

The distribution of European imports may be written as follows:

(3) MiT = θΤ Mi

Eσ3 (PiT / PiE)σ4

In this case, MiT stands for European volume import of commodity

(i) from Tunisia (or Tunisian exports to Europe).

Similarly, PiT represents the European import price index for

commodity (i) from Tunisia.

Elasticities verify the following conditions:

- Elasticity of imports from Tunisia compared to total imports: σ3>0

- Price elasticity of imports from Tunisia with respect to total

European imports, for product (i): σ4<0

At the theoretical level, the price elasticity of imports from Tunisia

(σ4) reflects the elasticity of substitution between imports from Tunisia

and imports from other countries. If imports from Tunisia are perfect


88


substitutes for imports from other countries, the price elasticity tends

to minus infinity, whereas, if they are complementary, the price

elasticity will tend towards zero.

1.1.3. Equation for Tunisian exports of product (i) to Europe

The two preceding equations express the European demand for

Tunisian commodities of sector (i):

(4) MiT = θΤ θΕ

σ3 Y σ3σ1 (PiE / Py ) σ3σ2 (PiT / PiE)σ4

This expression helps to show the:

- Income elasticity of imports of commodity (i) from Tunisia towards

Europe (σ3σ1>0)

- Price elasticity of imports of commodity (i) from Tunisia towards

Europe (σ4<0)

The estimation of the Equation for imports (i) from Tunisia to Europe

is realized by taking the logarithm of the above equation to obtain

a linear model:

(5) miT = χ0 y + β1 (piE - py) + β2 ( piT - piE ) + c

where:

miT = log(Mi

T) ; y = log(Y) ; py = log(Py) ; piE = log(PiE) ; piT = log(PiT)

χ0 = σ3σ1 ; β1 = σ3σ2 ; β2 = σ4 ; c= log(θΤ θΕσ3)

1.2 Estimating the Equation for Tunisian SectorExports: Error-Correction Model

1.2.1 Error-Correction Model

The econometric equation to be estimated will thus be as

follows:

(6) miT(t) = χ0 y(t) + β1 (piE(t) - py(t)) + β2 (piT(t) - piE(t)) + c + e(t)

where e(t) is the residual.

Hence, the estimated parameters of this equation will provide the

overall long-term elasticities compared to European income (χ0 > 0)

and with respect to Tunisian prices (a2<0).

Thus:

- a 1% increase in European income will be achieved in the long

run by an χ0% increase in Tunisian exports of commodity (i) to

Europe (ceteris paribus);

- a 1% increase in European prices will materialize in the long run

by a - β1% increase in Tunisian exports of commodity (i) to Europe

(ceteris paribus).

- a 1% increase in Tunisian prices will be achieved in the long run

by a - β2% decrease in Tunisian exports of commodity (i) to

Europe (ceteris paribus).

- a 1% increase in total European import prices will materialize in

the long run by a (β1 - β2)% variation in Tunisian export of

commodity (i) to Europe (ceteris paribus).

The estimation strategy of the model depends on the statistical

properties of the modelled series.

If one considers for example - which appears to be a reasonable

starting assumption to be confirmed by appropriate econometric

tests19 - that the variables of the model are I (1) using the terminology

of Engle and Granger (1987)20, i.e. if:

19 This involves implementing the usual unit root tests, e.g. Dickey and Fuller (1981).20 Intuitively, time series x(t) is integrated to order one (and noted I(1)) if it has a unit root, i.e. if it follows this process: x(t) = a.x(t-1) + b(L).Δx(t-1) + u(t) , where b(L) is a polynomialdelay, with a=1. Series x(t) will be I(0) if a<1.

89

miT(t) ~ I(1)

y(t) ~ I(1)

[piE(t) - py(t)] ~ I(1)

[piT(t) - piE(t)] ~ I(1)

the estimation of the model may be performed in two stages in

order to calculate long-term elasticities (by estimating the

cointegration relation 5) and short-term elasticities (by estimating

the error-correction model presented below).

The two-stage estimation is conducted as follows:

1. After verifying that the variables of interest are I(1), the following

cointegration relation (or long-term relation is estimated:

miT(t) = χ0 y(t) + β1 (piE(t) - py(t) ) + β2 ( piT(t) - piE(t) ) + c + ê(t)

2. The estimated residuals ê(t) are then tested and ascertained as

I(0) through the Engle and Granger (1981) cointegration tests.

3. If the previous stage validates the existence of a cointegration

relation, the error-correction model is then estimated as follows:

ΔmiT(t) = b1Δmi

T(t-1) + b2Δy(t) + b3 Δ[piE(t) - py(t)] + b4 Δ[piT(t) - piE(t)]

+ c0 + γ ê(t-1) + ε(t)

This expression helps to highlight the short-term value of income

elasticities (b2) and price elasticities (b3, b4), in addition to previously

estimated long-term elasticities.

The intuition behind this representation is as follows: Equation (5)

describes the "equilibrium" value of Tunisian imports, conditionally at

given income and relative prices. However, in the wake of a

shock affecting income or relative prices, Tunisian imports will not

immediately adjust but will converge more or less quickly towards

their new equilibrium. The adjustment speed towards the new

equilibrium is related to the value of parameter γ, (given that γ<0): if

γ is close to -1, the adjustment of imports towards the new equilibrium

will be long (thereby reflecting a strong inertia in imports relative to

European demand), whereas it will be faster if γ is close to 0.

The immediate effect of the shock affecting income or relative prices

is measured respectively by parameters b2, b3 and b4.

This representation may be used to simulate the expected exports

trajectories of Tunisia under different scenarios of change in relative

prices and income.

In conclusion, it is worth noting that if the two-stage approach helps

to intuitively present the distinction between short-term elasticities

and long-term elasticities, the one-stage estimation proposed by

Pesaran et al. (2001), supplemented by the critical values tabulated

for small samples by Narayan (2004), seems to be econometrically

more satisfactory and comprehensive than the two-stage approach

presented here.

1.2.2. Estimation of Dynamic Data Panel Models

Given that the stated objective is a sector-by-sector estimation of

price elasticities for Tunisian exports, the data panel estimation does

not seem to be relevant. However, if sector estimates turn out to be

disappointing (not robust or remote from theoretical a prioris), the

dynamic data panel estimation could be envisaged: either globally

or by a group of sectors with similar characteristics.


90


2. Econometric Estimation of Tunisian Sector Export

Equations

The modelling proposed in the previous section suggests three

potential explanatory variables of Tunisian exports:





However, preliminary estimates of export functions by sector have

revealed collinearity problems between different price indexes, which

made the estimated coefficients difficult to interpret. Therefore, the

following dynamic specification was retained, which is restricted to

the consideration of only one relative price (referenced p), which as

stated below, may take two different specifications:

(7) Δx(t) = ρΔx(t-1) + α0Δd(t) + α1Δd(t-1) + α2 Δd(t-2) + φ0Δp(t)

+ φ1Δp(t-1) + φ2Δp(t-2) + γ [ x(t-1) - βp(t-1) – d(t-1) + μ.t]

+ εt

Where:

x: logarithm of Tunisian exports to the European Union (EU) in

thousand Euros, constant 100= 2005 (compiled from comext).

d: logarithm of demand expressed to Tunisia by EU at constant

prices (compiled from comext). The expressed demand will be

presented in two alternative ways:

- The first considers demand expressed for all commodities in the

sector concerned (Agriculture, IAA,…);

- The second limits the demand to Tunisia's main exports in a

given sector (Agriculture, IAA,). Indeed, there is considerable

heterogeneity in the composition of commodities from each

sector, and in this form, expressed demand may be the more

representative model for Tunisian exports.

p: logarithm of the relative price of Tunisian exports. This relative

price is expressed in two alternative ways:

- The first one retains an index of sector relative prices for Tunisian

exports with respect to EU imports in the same sector;

- The second uses an index of relative prices for Tunisian exports

with respect to those of major competitors on the European market.

Hence, for each sector, 4 formulations were estimated according to

whether expressed demand should be calculated on all commodities or

restricted to the major export, and whether the relative price should be

calculated based on competitor prices or that of European Union countries.

It should be recalled that in this error-correction model (ECM), the term

in brackets [ x(t-1) - βp(t-1) – d(t-1)] represents the difference with respect

to long-term equilibrium wherein long-run elasticities are equal to 1 with

regard to expressed demand (in fact, Tunisia's market share trend is

modelled on the European market) and β in relation to relative price.

The term μ.t stands for the potential deterministic market share trend

which is growing annually at μ rate.

Lastly, the variables Δx represent the variation of x, and dummy

variables were sometimes included in the regression model.

2.1 Data

The estimation period spans from 1988 to 2008 in annual data. The

database was built within the ITCEQ team and comprises 8 sectors

91

as follows:

Sector 1: Agriculture (Agr)

Sector 2: Agri-food industry (IAA)

Sector 3: Ceramic Building Materials and Glass (MCCV)





Sector 8: Hydrocarbons (Hyd).


The method of estimation selected here is an Error-Correction Model

(ECM) estimated in one stage, following the approach proposed by

Pesaran et al. (2001). It involves estimating an unconstrained form

of equation (6):

Δx(t) = ρΔx(t-1) + α0Δd(t) + α1Δd(t-1) + α2 Δd(t-2) + φ0Δp(t) +

φ1Δp(t-1) + φ2Δp(t-2) + a.x(t-1) + b.p(t-1) + c.d(t-1) + εt

and to test the significance of parameter (a): if it is significantly different

from zero, it is concluded that there is an error-correction mechanism

and the long-term coefficients can easily be determined from

estimated parameters a, b and c.

Before proceeding with the econometric estimation, the order of

integration of the series used was tested using the ADF test. The

results show that the entire series tested are all integrated to order 1.

2.3 Key Outputs

2.3.1. Agriculture

The key outputs, for the 4 specifications tested for this sector, are

summarized in Table 1:

• The value of the long-run elasticity of demand for exports was

imposed to unity. It is worth noting that, in preliminary tests, this

restriction is more easily accepted by using expressed demand

calculated on the basis of major products;

• Whenever expressed demand for "Total commodities " is used,

our estimates reveal a significantly positive trend, representing

an increase in market share trend for this sector;

• Long-term price elasticities bear the expected sign in 3 out of 4

cases. Only the formulation with EU prices and "major

commodities" for demand bears a positive sign, contrary to

economic intuition. The value of this elasticity is low, to the tune

of -.2 in the case of competitor prices and -0.3 in that of EU

countries, which would imply that Tunisian products in the sector

have few substitutes on the EU import market; and

• ECMs that use "Total commodities" for demand have satisfactory

statistical properties. LM tests result in the rejection of the hypothesis

of autocorrelation in the residuals of these equations. These residuals

are homoskedastic under the White test and the ARCH test. The

functional form of the equation passed the Reset test. Lastly,

according to the Jarque Bera test, the residuals of all equations

are normally distributed. That is not the case for those using "Major

commodities " for demand.


92


At the end of these estimates, it seems the best equation is the one

that uses "Total exports" as expressed demand and considers the

prices of EU countries as foreign price.

Graph 1 illustrates the static adjustment of this ECM in the past.


LT Elasticities


Buoyancy (-3.70) (-1.45) (-2.21) (-3.28) γ

Relative prices of …

… competitors -0.20 (-0.28) -0.17 (-1.00) β

… EU as a whole -0.30 (-2.09) 0.06 (0.12)

Trend 0.007 (2.12) 0.02 (1.81) 2.0 2.0 μ

ST Elasticities

Δlog(expressed demand) 1.49 (3.57) α0

Dummy

d98 0.16** (1.91) 0.16* (2.03) η0

d06 0.23 (2.32) η1

d02 -0.14 (-1.72) η2

Tests

LM (2) 0.82 0.21 0.03 0.10

Arch(1) 0.72 0.33 0.72 0.33

Normality 0.71 0.56 0.92 0.38

Reset (2) 0.97 0.29 0.08 0.05

R-squared 0.63 0.50 0.44 0.39


Ranking 1 4 3 2

Source: ITCEQ/DEFI CalculationsNote: Student's t-statistic is shown in brackets ne: coefficient was not estimated but imposed.

Table 1: Summary of Key Findings of the Export Equation Estimates for Sector 1 (Estimation Period 1988-2008)

93


Graph 1: Annual Growth rate of Tunisian Exports in Sector 1 observed and adjusted

Source : ITCEQ/DEFI Calculations

Graph 2 represents the response of Agriculture sector exports in

the wake of 2 simulated shocks, by retaining the best specification

from Table 1 (column 1):

- A 1% increase in expressed demand;

- A 1% increase in the relative price of Tunisian exports.

This graph shows that the median adjustment period21 is slightly

above one year in the case of a price shock, whereas in the case

of a demand shock, there currently is a phenomenon of short-term

"overshooting" versus long-term equilibrium (elasticity of demand

stands at 1.49 in the short run and at 1 in the long term.

These results imply that following a decline in demand for imports

in the EU21, Tunisian agricultural exports may be substantially

affected (about a 1.5% decline, with respect to the trend

throughout the year, in Tunisian exports in case of a 1% drop in

European demand), which would make it one of the sectors most

affected by the European economic downturn. Furthermore, the

consequence of the low price elasticity is that a fall in Tunisian

prices would not be able to significantly offset the decrease in

exports during a recession in Europe.

21 C’est-à-dire le nombre d’années nécessaires pour que la moitié de l’ajustement total soit réalisé.

94


A 1% increase in expressed demand A 1% increase in the relative price of Tunisian exports

Graph 2 : Responses to shocks

Source : Calculs ITCEQ/DEFI

2.3.2. IAA Sector

The key outputs for this sector are summarized in Table 2:


unity. It is worth noting that this restriction is accepted in all cases;

• Long-term price elasticities have the expected sign in three out

of four cases. Only the formulation with the price of EU countries

and "major commodities " for demand bears a positive sign,

contrary to economic intuition. The value of this elasticity is -0.7

in the case of competitor prices and -0.5 in that of EU countries;

and

• All ECMs have satisfactory statistical properties. LM tests

result in the rejection of the hypothesis of autocorrelation in

the residuals of these equations. These residuals are

homoskedastic under the White test and the ARCH test. The



are normally distributed.

According to these estimates, the best equation is the one that uses

"Total commodities" for demand and considers the prices of EU

countries as foreign price. Findings are summarized in the table

below.

95



LT Elasticities


Buoyancy -1.13 (-6.10) -1.19 (-5.82)) -1.19 (-9.94) -1.33 (-5.99) γ


… competitors -0.64 (-1.87) -0.74 (-2.39) β

… EU as a whole -0.54 (-1.61) 0.26 (2.47)

Trend μ

ST Elasticities

Δlog(expressed demand) -0.89 (-2.61) -0.93 (-2.54) -0.75* (-2.06) α0

Dummy

d02 -0.93 (-3.52) -0.87 (-3.25) -0.98 (-3.97) -1.12 (-4.11) η0

d01 -0.96 (-3.10) -0.90 (-2.74) -0.93 (-3.13) -1.15 (-3.32) η1

d96 -0.14 (-1.72) -0.79 (-3.32) η2

d94 0.48* (2.09) 0.62 (3.74) η3

Tests

LM (2) 0.55 0.32 0.37 0.87

Arch(1) 0.85 0.70 0.33 0.93

Normality 0.80 0.86 0.83 0.74

Reset (2) 0.10 0.52 0.12 0.76

R-squared 0.87 0.81 0.86 0.82


Ranking 1 3 4 2



Graph 3 illustrates the static adjustment of this ECM in the past.

96


Graph 3: Annual Growth rate of Tunisian Exports in Sector 2 observed and adjusted


Graph 4 represents the response of IAA sector exports in the wake

of a 1 % shock on demand and relative price, by retaining the best

specification from Table 2 (column 1).

The median adjustment period is very short, i.e. less than one year,

in the case of a demand shock, whereas, in the case of a price shock,

there is short-term overshooting with respect to long-term equilibrium

(price elasticity stands at -0.89 in the short run and at -0.54 in the

long term). Hence, a recession in Europe would have very swift and

egregious adverse impacts on Tunisian IAA sector exports. However,

a fall in export prices would be able to, at least partly offset such an

impact.



Graph 4: Responses to Shocks

97


2.3.3. Ceramic Building Materials and Glass Sector (MCCV)


• The value of elasticity of exports demand was imposed to unity.

It is worth noting that this restriction is accepted econometrically

in the case of "Total commodities ". The elasticity in the case of

"Major commodities " is close to 0.6 when the coefficient is

left free.

• Long-term price elasticities bear the expected sign in all cases.

The value of this elasticity is close to -0.5.

• ECMs that use "Total commodities " for demand have

satisfactory statistical properties. LM tests result in the rejection

of the hypothesis of autocorrelation in the residuals of

these equations. These residuals are homoskedastic under

the White test and the ARCH test. The functional form of the

equation passed the Reset test. Lastly, according to the Jarque


That is not the case for those using "Major commodities " for

demand.


LT Elasticities


Buoyancy -0.88 (-6.66) -0.87 (-6.63) -0.74 (-4.70) -0.73 (-4.92)γ


… competitors -0.46 (-3.81) -0.57 (-3.21)β

… EU as a whole -0.44 (-3.64) -0.49 (-2.69)

Trend μ

ST Elasticities

Dummy

d9293 -0.52 (-2.86) -0.51 (-2.84) -0.68 (-2.78) -0.63 (-2.76)η0

Tests

LM (2) 0.52 0.70 0.49 0.61

Arch(1) 0.35 0.41 0.58 0.75

Normality 0.63 0.68 0.26 0.23

Reset (2) 0.37 0.23 0.01 0.01

R-squared 0.85 0.86 0.76 0.79


Ranking 1bis 1 3bis 3



The best equation is one that uses "Total exports" as expressed

demand. (1st column). The distinction on prices does not seem to

be differentiating in this sector.

Graph 5 illustrates the static adjustment on the past of this ECM.

98


In the graph for responses to shocks (Graph 6 based on Table 3,

column 1), the median period is short, about a year, for both types

of shocks: a 1% fall in European demand should result in about a

0.5% decrease in IMCCV sector exports over one year.

Source: ITCEQ/DEFI Calculations

Graph 5: Annual Growth Rate of Tunisian Exports in Sector 3 observed and adjusted



Graph 6 : Responses to Shocks

99


2.3.4. Mechanical and Electrical Industries Sector




around 1.3 in both cases;

• In all ECMs, estimates reveal a significantly positive trend,

signifying an increased market share trend in this sector. This is

consistent with elasticity of expressed demand exceeding unity;

• Long-term price elasticities do not have the expected sign when

competitor prices are included in the relative price. In the other

case, the estimated elasticity is relatively high, between -1 and

-1.5; and









LT Elasticities


Buoyancy -0.22 (-2.31) -0.13 (-2.27) -0.66 (-8.86) -0.73 (-2.71) γ


… competitors 0.09 0.27 β

… EU as a whole -1.10 -1.56

Trend 0.09 0.08 0.10 0.06 μ

ST Elasticities

Δlog(expressed demand) 0.42 (4.36) 0.35 (2.92) 0.72 (8.91) α0

Δ(Relative prices) -0.32 (-2.77) -0.14 (-1.24) -0.53 (-6.24) α0

Δ(Relative prices) -1 0.39 (4.12) α1

Δlog(Exports)-2 -0.28 (-4.91) ρ0

Dummy

d89 0.14 (3.60) 0.12 (2.48) η0

d06 0.16 (3.28) 0.17 (2.70) 0.10 (7.65) η1

d96 -0.18 (-3.68) η2

Tests

LM (2) 0.34 0.54 0.36 0.51

Arch(1) 0.72 0.86 0.16 0.32

Normality 0.67 0.34 0.72 0.35

Reset (2) 0.64 0.40 0.83 0.35

R-squared 0.83 0.75 0.98 0.63


Ranking 1bis 3 1 3bis

Table 4: Summary of Key Findings of the Export Equation Estimates for Sector 4 (Estimation Period: 1988-2008)


100


The best equation is one that uses prices calculated on all EU countries

in the competitiveness indicator, and for major exports only.




The response to simulated shocks is shown in Graph 8 by retaining

the best specification from Table 4 (column 3). For the demand shock,

the median period is instantaneous (given that short-term elasticity is

greater than 0.5 for a long-term elasticity equal to 1), and it is less

than one year for the relative price shock. The effects of a European

recession here are similar to those of IMCCV sector: a 1% drop in

European demand should lead to a rapid decrease of about 0.5% (in

deviation from trend) in IME sector exports. However, the sensitivity

of these exports to relative prices implies that a reduction in prices

would be able to partially offset this movement.

101




2.3.5. Chemical Sector




around 1.5 in all cases, except for case 4 (expressed demand,

“major commodities”, competitor prices), where elasticity stands

at 0.8;

• In all ECMs, estimates reveal a significantly negative trend,

signifying a loss-of-market share trend in this sector. This is

consistent with elasticity of demand below unity;


formulation used, although their value varies greatly: it stands

around -0.7 / -0.9 in the case of a "Total commodities" demand.

These values become much higher (-1.89 and -4.67 respectively)

when expressed demand includes "Major commodities". This

indicates a high degree of substitutability of Tunisia's chemical

exports with respect to competing exports; and


result in the rejection of the hypothesis of autocorrelation

in the residuals of these equations. These residuals are

homoskedastic under the White test and the ARCH test. Lastly,


are normally distributed. The functional form of the equation

alone failed the Reset test.

102



LT Elasticities


Buoyancy -0.75 (-2.70) -0.67 (-2.36) -0.23 (-1.99) -0.39 (-3.55) γ


… competitors -0.88 (-2.35) -4.67 (-3.91) β

… EU as a whole -0.70 (-2.69) -1.89 (-2.09)

Trend -0.06 (-3.40) -0.06 (-3.19) -0.11 (-2.63) -0.13 (-4.47) μ

ST Elasticities

Δlog(expressed demand) 0.87 * (2.01) 0.72 ** (1.65) α0

Δ(Relative prices) -0.32 (-2.77) -0.14 (-1.24) -0.15 ** (-1.79) α0

Δ(Relative prices) -1 1.52 (2.90) φ1

Δlog(Exports)-2 1.39 (3.00) φ2

Dummy

d9394 -0.18* (-2.26) -0.19 * (-2.28) -0.24 (-2.69) -0.28 (-4.08) η0

Tests

LM (2) 0.75 0.42 0.73 0.19

Arch(1) 0.49 0.91 0.37 0.30

Normality 0.63 0.68 0.61 0.42

Reset (2) 0.07 0.10 0.07 0.02

R-squared 0.74 0.73 0.65 0.86


Ranking 2 3 4 1



At the end of these estimates, it seems the best equation is one that

uses prices calculated based on competitor prices in thecompetitiveness

indicator and considers expressed demand as "Total commodities".


103




Source : Calculs ITCEQ/DEFI


Chemical sectorChemical sector


The median adjustment period for chemical industry exports seems

to be long (Graph 10): to the tune of 2 years for demand shock and 4

years for relative price shock. In the scenario of a European recession,

exports should not be excessively reduced (in deviation from trend)

over a period of one year, whereas high price elasticity helps to consider

relative price adjustments capable of offsetting such reduction.

104


2.3.6. Textile, Apparel and Leather Sector



unity. Such unstressed elasticity would fall below unity, and stand

at about 0.6 in all cases;


formulation used, and their value is high, standing between -3

and -4 when competitor prices are used and between -5 and -

8 in the other case. Consequently, as in the case of chemical

industries, there would be a phenomenon of high substitutability

of these exports on the European market; and


result in the rejection of the hypothesis of autocorrelation

in the residuals of these equations. These residuals are






LT Elasticities


Buoyancy -0.14 (-2.12) -0.15 (-3.31) -0.10 (-2.21) -0.15 (-3.56) γ


… competitors 2.88 (-9.01) -4.21 (-8.27) β

… EU as a whole -5.30 (-4.99) -8.04 (-6.39)

Trend μ

ST Elasticities

Δ(Relative prices) -0.26 (-2.33) -0.34 (-3.10) -0.43 (-3.63) -0.46 (-3.94) φ0

Dummy

d9394 0.16 (2.41) 0.15 (2.25) 0.19 (2.73) η0

Tests

LM (2) 0.84 0.48 0.62 0.46

Arch(1) 0.36 0.69 0.62 0.88

Normality 0.69 0.98 0.64 0.58

Reset (2) 0.16 0.89 0.77 0.77

R-squared 0.65 0.84 0.80 0.86


Ranking 4 2 3 1



105

At the end of this initial study, it would seem the best equation is

one that uses prices calculated based on competitor prices in the

competitiveness indicator and considers expressed demand as

"Total commodities".

Graph 11 illustrates the static adjustment on the past of this

ECM.


The estimated median adjustment periods for Textiles, Apparel

and Leather industry exports are the longest among the sectors

studied (Figure 12): they are close to 5 years for the demand shock

and over 3 years for the relative price shock. In the short term, a

European recession should not have much effect on those exports.

As in the case of the chemical industry, significant price elasticity

helps to consider relative price adjustments capable of offsetting

a possible decline in exports.





Textile, Apparel and LeatherTextile, Apparel and LeatherGraph 12: Responses to Shocks

106

2.3.7. Miscellaneous Sector


• The value of elasticity of demand addressed for exports was

imposed to unity. It is worth noting that this restriction is accepted

more easily by using expressed demand calculated on the basis

of major exports (estimated elasticity of 1.05) than for "Total

commodities" (estimated elasticity greater than 1.7).

• Price elasticities never have the expected sign except for

case 3 (expressed demand for "Major commodities" and

prices of EU countries. In the latter case, the estimated value

is -1.5.






according to the Jarque Bera test, the residuals of all

equations are normally distributed.



LT Elasticities


Buoyancy -0.12 (-2.59) -0.31 (-2.96) -0.38 (-4.62) -0.12 (-2.18)γ


… competitors 0.07 (0.30) 0.41 (0.69) β

… EU as a whole 2.52 (2.71)-1.43 (-3.74)

Trend 0.08 (2.57)0.05 (3.74)

0.07 (1.46) μ

ST Elasticities

Δ(Relative prices) 0.09 ** (1.49)

α0

Δlog(Expressed demand)-10.16 (3.00)

α1

Dummy

d98 0.18 (2.42)0.18 (2.14) η0

d02 0.19 * (2.00)0.19 (2.22)

0.20 (2.66)η1

Tests

LM (2) 0.37 0.60 0.74 0.36

Arch(1) 0.74 0.75 0.78 0.82

Normality 0.90 0.08 0.63 0.03

Reset (2) 0.65 0.96 0.64 0.95

R-squared 0.43 0.60 0.71 0.66


Ranking 4 3 1 2



107


The best equation is one that uses prices calculated based on

competitor prices in the competitiveness indicator and considers

expressed demand as "Total commodities".






Miscellaneous SectorMiscellaneous SectorGraph 14: Responses to Shocks

The estimated responses to shocks for this sector (Graph 14)

reveal very short median adjustment periods (about 1 year) following

a demand shock, and to the tune of 2 years in the wake of a price

shock. Hence, a 1% fall in European demand would be attended

by a 0.5% drop in sector exports within 1 year (in deviation

from trend).

108


2.3.8. Hydrocarbons and Refined Products Sector





commodities” (estimated elasticity is 1.03) than in the case of


• Price elasticities always have the expected sign. The value ranges

between -8 in the case of EU prices and around -5 in the case

of competitor prices. This sector is characterized by the highest

price elasticities among the sectors studied. This is not surprising,

given the type of products involved in the sector: hydrocarbons

are standardized exports, and Tunisian products are close

substitutes with respect to exports traded on the European

market; and


in the rejection of the hypothesis of autocorrelation in the residuals

of these equations. These residuals are homoskedastic under

the White test and the ARCH test. The functional form of the

equation passed the Reset test. Lastly, according to the Jarque



LT Elasticities


Buoyancy -0.83 (-6.31) -0.41 (-2.31) -0.66 (-3.75) -0.38 (-2.02)γ


… competitors -5.65 (-1.69) -4.47 (-1.29)β

… EU as a whole -7.93-7.93 (-3.81)

Trend μ

ST Elasticities

Dummy

d9394 -0.53 (-4.52) -0.68 (-3.60) -0.65 (-5.09) -0.71 (-3.74)η0

Tests

LM (2) 0.11 0.26 0.65 0.30

Arch(1) 0.83 0.84 0.12 0.84

Normality 0.89 0.27 0.73 0.21

Reset (2) 0.80 0.68 0.30 0.62

R-squared 0.89 0.69 0.86 0.67


Ranking 1 3 2 4



109


At the end of this initial work, it seems the best equation is one that

uses "Total commodities" as expressed demand and considers the

prices of EU countries as foreign price.






hydrocarbon and Refined Products Sectorhydrocarbon and Refined Products SectorGraph 16: Responses to Shocks

As might be expected, median adjustment periods are very

short: less than one year following demand shocks or price

shocks. The notable fact here is that high price elasticity

and short adjustment periods jointly make exports from this

sector very sensitive to short-term fluctuations in relative

prices.

110


2.3.9. Total Goods





commodities” (estimated elasticity is 1.01) than in the case of


• Price elasticities always have the expected sign except for

formulation 3 (prices of all EU countries and expressed demand

as “Major commodities”). In the other cases, the value stands

between -0.3 and -0.5; and





functional form of the equation passed the Reset test.

Lastly,Bera test, the residuals of all equations are normally

distributed.


LT Elasticities


Buoyancy -0.53 (-5.14) -0.49 (-4.28) -0.53 (-3.93) -0.57 (-5.59) γ


… competitors -0.49 (-3.30) -0.31 (-2.40) β

… EU as a whole -0.53 (-2.18) 0.35 (1.05)

ST Elasticities

Δlog(Expressed demand)-1 -0.44 (-2.65) α1

Δ(Relative prices)-1 -0.35 (-1.95) φ1

Dummy

d9394 -0.14 (-4.89) -0.11 (-5.07) -0.13 (-4.75) -0.12 (-4.94) η0

Tests

LM (2) 0.59 0.36 0.78 0.87

Arch(1) 0.55 0.86 0.51 0.43

Normality 0.73 0.96 0.20 0.42

Reset (2) 0.08 0.17 0.14 0.87

R-squared 0.78 0.83 0.75 0.82


Ranking 3 2 4 1



111



Graph 17: Annual growth rate of Tunisian exports to Total Assets observed and adjusted

The best equation is one that uses "Total commodities" as expressed

demand and considers competitor prices as foreign price.

Graph 17 illustrates the static adjustment on the past of this

ECM.

Graph 18 is used to summarize the dynamic effects that could

be expected from a 1% decline in European demand for the

volume of Tunisian exports. Given that these are mainly performed

in the Textiles, Apparel and Leather sector (approximately 40%

of total exports of goods in 2008 in the sector coverage of the

database used) and Mechanical and Electrical Industries (about

30% of total exports of goods in 2008, the same sources),

responses to shocks are closely linked to the dynamic behaviour

of both sectors.

Regarding the demand shock, the median adjustment period is

about one year: a 1% drop in European demand would therefore

materialize by reducing the volume of goods exported by about

0.5% over one year.

Long-term price elasticity is relatively low (-0.31) and the median

period of adjustment to a price shock is significantly greater than

one year: overall, a limited variation in relative prices should have

little impact on the quantities exported over one year.

112



Total AssetsTotal Assets A 1% increase in expressed demand A 1% increase in the relative price of Tunisian exports


2.3.10. Summary

Table 10 summarizes the ranking of formulations by sector. It seems

that formulation 1 (expressed demand as "total commodities" and

prices of all EU countries as foreign prices) appears as the most

acceptable if a formulation common to all sectors should be retained.

Relative prices of… Total Commodities Major Commodities

Sector 1… competitors 4 2

… EU as a whole 1 3



Sector 3… competitors 1

… EU as a whole 1 bis 3 bis


… EU as a whole 1 bis 1









Total Goods… competitors 2 1


Table 10: Summary of Best Sector Specification


113

In conclusion, the findings of the estimates conducted provide

important lessons, as they highlight huge differences among sectors,

both with respect to the long-term price elasticities of the various

sectors (and hence their level of substitutability on the European

market), and to the dynamic behaviour of Tunisian exports in the

wake of relative price shocks and demand shocks.

It should be emphasized that responses to shocks were discussed

with the example of a European recession (and a decrease in import

demand from the EU). However this does not, in any way, involve

short-term economic forecasting: indeed, while it is true that

European imports plummeted (to the tune of 40% in value and in

monthly data) from their peak in July 2008 to record lows in

September 2009, it should be acknowledged also that they have,

since that date, recorded very strong growth and are almost back

to their pre-crisis level since January 2011.


114


3. Econometric Estimates of Tunisia's Sector Export

Equations through Dynamic Panel Analysis

Export functions were estimated through dynamic panel

analysis.

3.1 Data

The estimation period spans from 1988 to 2008. The database was

built with the ITCEQ team and comprises 8 sectors presented in the

previous section:



Sector 3: Agri-food Industry (IAA)

Sector 4: Ceramic Building Materials and Glass Industry (IMCCV)



Sector 7: Agriculture (AGRICULTURE)

Sector 8: Hydrocarbons (HYDRPDSRAFFINES)

Panel data variables are as follows:

X: Tunisian exports, the logarithm will be denoted by LX and the

logarithm variation will be denoted by DLX;

D: European demand, the logarithm will be denoted by LD and the

logarithm variation will be denoted by DLD;

PCON: Relative price of Tunisian exports with respect to competitor

prices, the logarithm will be denoted by LPCON and the logarithm

variation will be denoted by DLPCON;

PUE: Relative price of Tunisian exports with respect to European

Union prices, the logarithm will be denoted by LPUE and the logarithm

variation will be denoted by DLPUE

The modelled variables will be expressed in logarithms and before

proceeding with the econometric estimation proper, the order of

integration of the series used was tested with the panel unit root

tests.

Table 11: Findings of Panel Unit Root Tests

Panel unit root test: Summary Sample: 1988 2008

Exogenous variables: Individual effects, individual linear trends

Automatic selection of maximum lags based on SIC: 0 to 3

Newey-West automatic bandwidth selection and Bartlett kernel

Tunisian Exports (LX)

Method Statistic Prob.** Cross-sections Obs

Null: Unit root (assumes common unit root process)

Levin, Lin & Chu t* -1.58336 0.0567 8 156

Breitung t-stat 3.92219 1.0000 8 148

Null: Unit root (assumes individual unit root process)

Im, Pesaran and Shin W-stat -1.94229 0.0261 8 156

ADF - Fisher Chi-square 31.8442 0.0105 8 156

PP - Fisher Chi-square 42.1884 0.0004 8 160

115

The findings of unit-root tests differ from one another and indicate

that the series may be stationary or integrated to order 1.

In the case the series are assumed to be integrated to order 1.

Pedroni's Cointegration tests (Table 12) also give conflicting results

(the null hypothesis of noncointegration may be rejected or not

rejected by the test applied).


Demand (LD)



Levin, Lin & Chu t* -1.46385 0.0716 8 152

Breitung t-stat -0.29669 0.3834 8 144





Relative prices calculated based on competitor prices (LPCON)



Levin, Lin & Chu t* -2.09402 0.0181 8 155

Breitung t-stat 0.84405 0.8007 8 147





Relative prices calculated based on European Union prices (LPUE)



Levin, Lin & Chu t* -3.55244 0.0002 8 151

Breitung t-stat 1.31856 0.9063 8 143





** Probabilities for Fisher tests are computed using an asymptotic Chi -square distribution. All other tests assume asymptotic normality.

116


Table 12: Findings of panel cointegration tests

Series: LX - LD - LPUE Sample: 1988 2008 Included observations: 168 Cross-sections included: 8

Null Hypothesis: No cointegration

Automatic lag length selection based on SIC with a max lag of 3

Newey-West automatic bandwidth selection and Bartlett kernel

Pedroni’s Test (Residual Cointegration Test) Trend assumption: Deterministic intercept and trend

Alternative hypothesis: common AR coefs. (within-dimension)

Statistic Prob. Weighted Statistic Prob.

Panel v-Statistic 0.714735 0.2374 0.116556 0.4536

Panel rho-Statistic -0.417653 0.3381 0.240157 0.5949

Panel PP-Statistic -3.395447 0.0003 -2.978726 0.0014

Panel ADF-Statistic -3.452005 0.0003 -3.052121 0.0011


Statistic Prob

Group rho-Statistic 0.851300 0.8027

Group PP-Statistic -2.709311 0.0034

Group ADF-Statistic -2.610142 0.0045

Kao’s Test (Residual Cointegration Test) Trend assumption: No deterministic trend


Statistic Prob.

ADF -1.627718 0.0518

Residual variance 0.061223

HAC variance 0.031124

Johansen Test (Fisher Panel Cointegration Test) Trend assumption: Linear deterministic trend

Lags interval (in first differences): 1 1

Unrestricted Cointegration Rank Test (Trace and Maximum Eigenvalue)

HypothesizedNo. of CE(s)

Fisher Stat.*(from trace test) Prob.

Fisher Stat.*(from max-eigen test) Prob.

None 51.44 0.0000 38.41 0.0013

At most 1 27.00 0.0415 24.72 0.0750

At most 2 21.45 0.1618 21.45 0.1618

* Probabilities are computed using asymptotic Chi-square distribution.

117


The following specifications will be estimated hereafter:

- First, an autoregressive distributed lag (ADL) model in level will

be estimated (assuming stationarity of the series), by using two

alternative specifications for relative prices: one based on prices

in the EU and that based on competitor prices on the European

market;

- Second, an autoregressive distributed lag (ADL) model will be

estimated in first differences (assuming non-stationarity of the

series), by using the two alternative specifications for relative

prices; and.

- Lastly, an -error-correction model (ECM) will be estimated in one

step, which helps to include integrated and non-cointegrated

cases, still under the two alternative specifications of relative prices.

3.3 Key Outputs

3.3.1. ADL model in level in the case of EU prices (LPUE)

The key outcomes of the panel estimation are summarized in Table 8:

• The estimated value of elasticity of demand for exports stands

at 0.82 in the long run and 0.27 in the short run.

• A significantly positive trend ensues from these estimations,

which represents a market share growth trend.

• Price elasticities bear the expected sign: the value of such

elasticity stands at -0.71 in the short run and long run.

• The ADL model has satisfactory statistical properties.


Table 13: Estimation of Exports in Level in the Case of EU Prices

Dependent Variable: LX

Method: Panel Least Squares

Sample (adjusted): 1989-2008

Periods included: 20

Cross-sections included: 8

Total panel (balanced) observations: 160

Cross-section SUR (PCSE) standard errors & covariance (d.f. corrected)

Variable Coefficient Std. Error t-Statistic Prob.Constant 3.679917 0.729487 5.044528 0.0000

LX(-1) 0.677151 0.071589 9.458823 0.0000

LD 0.265131 0.096423 2.749664 0.0067

LPUE -0.705671 0.079139 -8.916812 0.0000

LPUE(-1) 0.475012 0.087979 5.399130 0.0000

Trend 0.037081 0.008620 4.301492 0.0000

Effects SpecificationCross-section fixed (dummy variables)

R-squared 0.976254 Mean dependent var 12.44637

Adjusted R-squared 0.974316 S.D. dependent var 1.413753

S.E. of regression 0.226571 Akaike info criterion -0.053759

Sum squared resid 7.546150 Schwarz criterion 0.196099

Log likelihood 17.30074 Hannan-Quinn criter. 0.047699

F-statistic 503.6370 Durbin-Watson stat 2.302661

Prob(F-statistic) 0.000000


118


Graph 18 below illustrates the static adjustment on the past of this ADL model.

3.3.2. ADL model in levels in the case of competitor prices(LPCON)

The key outcomes for this model are summarized in Table 14:

• The value of elasticity of demand for exports stands at 0.83

in the long run and 0.26 in the short run.

• A significantly positive trend ensues from these estimations,

which represents a market share growth trend in this sector.

• Estimated price elasticities stand at -0.65 in the short run and

-0.71 in the long run.

• The statistical properties of the residuals of this equation are

satisfactory.

Hence, there is little difference between both specifications: that

which retains EU prices in the relative price and that which uses

competitor prices on the European market.


Graph 18: Tunisian Exports observed and adjusted (in log)

119

Graph 19 illustrates the static adjustment on the past of this ADL model.


Table 14: Level export estimates in the case of competitor prices

Dependent Variable: LX


Sample (adjusted): 1989-2008





Variable Coefficient Std. Error t-Statistic Prob.Constant 3.532320 0.689397 5.123782 0.0000LX(-1) 0.688561 0.073388 9.382525 0.0000LD 0.258474 0.099060 2.609279 0.0100LPUE -0.645157 0.079857 -8.078912 0.0000LPUE(-1) 0.423191 0.084779 4.991701 0.0000Trend 0.025265 0.008747 2.888411 0.0045

Effects SpecificationCross-section fixed (dummy variables)R-squared 0.974836 Mean dependent var 12.44637Adjusted R-squared 0.972782 S.D. dependent var 1.413753S.E. of regression 0.233240 Akaike info criterion 0.004265Sum squared resid 7.996959 Schwarz criterion 0.254122Log likelihood 12.65883 Hannan-Quinn criter. 0.105723F-statistic 474.5551 Durbin-Watson stat 2.356153Prob(F-statistic) 0.000000



Graph 19: Tunisian exports observed and adjusted (in log)

120


3.3.3. ADL model in first differences in the case of EUprices (DLPUE)


• The export demand variable is not statistically significant.

• Price elasticities bear the expected sign, and the value of this

elasticity stands at -0.67 in the short run and -0.70 in the long run.

• This model has satisfactory statistical properties. However,

this specification seems to be less relevant on account of the

non-significant value of the demand parameter.

Hence, while it is true that this model is in principle less relevant than

the previous ones, it is worth noting also that the value of price elasticities

is of a comparable order of magnitude, both in the long and short terms.

Table 15: Export equation estimates of growth rates in the case of EU prices

Dependent Variable: DLX


Sample (adjusted): 1990- 2008





Variable Coefficient Std. Error t-Statistic Prob.

Constant 0.064636 0.022456 2.878397 0.0046

LX(-1) -0.310827 0.125930 -2.468259 0.0148

DLD 0.311570 0.210593 1.479487 0.1413

DLPUE -0.674955 0.084278 -8.008625 0.0000

DLPUE(-1) -0.237546 0.107686 -2.205923 0.0290




S.E. of regression 0.238345 Akaike info criterion 0.045462






121



3.3.4. ADL model in first differences in the case of competitor

prices (DLPCON)


• The export demand variable is not statistically significant.



• The estimated model has satisfactory statistical properties.

However, this specification, as in the previous case, seems to

be less relevant than level estimates on account of the non-

significant value of the demand parameter.


Graph 20: Annual Growth Rate of Tunisian Exports observed and adjusted

122


Table 16: Estimating Exports Equations (ECM) in the case of EU prices








Variable Coefficient Std. Error t-Statistic Prob.Constant 0.071608 0.022456 2.878397 0.0053LX(-1) -0.321674 0.125930 -2.468259 0.0136DLD 0.104188 0.210593 1.479487 0.6341DLPCON -0.581546 0.084278 -8.008625 0.0000DLPCON(-1)) -0.238305 0.107686 -2.205923 0.0212Effects SpecificationCross-section fixed (dummy variables)R-squared 0.426061 Mean dependent var 0.052422Adjusted R-squared 0.380966 S.D. dependent var 0.313173S.E. of regression 0.246401 Akaike info criterion 0.111940Sum squared resid 8.499850 Schwarz criterion 0.350667Log likelihood 3.492579 Hannan-Quinn criter. 0.208919F-statistic 9.448029 Durbin-Watson stat 2.141201Prob(F-statistic) 0.000000





123


3.3.5. Error-Correction Model (ECM) in the case of EUPrices (LPUE)


• The export demand variable is not statistically significant in

the short term. The value of elasticity of demand for exports

stands at 0.98 in the short run, and is not significantly different

from the unit value.


elasticity stands at -0.73 in the short run and -0.61 in thelong run.

• The ECM has satisfactory statistical properties.









Constant 2.280052 0.556009 4.100751 0.0001

DLPUE -0.729278 0.086822 -8.399715 0.0000

DLD 0.360015 0.215213 1.672830 0.0965

LX(-1)-LD(-1)-LPUE(-1) -0.219796 0.066134 -3.323475 0.0011

LD(-1) 0.020319 0.054852 0.370431 0.7116

LPUE(-1) -0.385218 0.118384 -3.253987 0.0014











124




Graph 22: Annual Growth Rates of Tunisian Exports observed and adjusted

3.3.6. Error-Correction Model (ECM) in the case ofCompetitor Prices (LPCON)


• The export demand variable is not statistically significant in

the short term. The value of elasticity of demand for exports

stands at 0.99 in the short run, and is not significantly different

from unity.



• The ECM has satisfactory statistical properties.

125










Constant 2.982117 0.652054 4.573424 0.0000

DLPCON -0.673427 0.080996 -8.314298 0.0000

DLD 0.136647 0.214470 0.637140 0.5250

LX(-1)-LD(-1)-LPUE(-1) -0.260304 0.068674 -3.790439 0.0002

LD(-1) 0.014654 0.058726 0.249525 0.8033

LPCON(-1) -0.511016 0.122944 -4.156494 0.0001











126


In conclusion, the data panel estimates helped to meaningfully

supplement the estimates made for each sector considered separately.


panel estimation allows for a rather robust estimation of the 'average'

dynamic behaviour of sector exports. It follows that, whatever the

specifications used for the measurement of relative prices:

- The elasticity of Tunisian exports to Europe in relation to expressed







substantially greater than one year;

- The price elasticity of Tunisian exports to Europe stands at -0.6

to -0.73 both in the short and long terms;

- Lastly, the residual plots of the various models estimated suggest



the sector approach developed in the previous section. Indeed,

it has been revealed in this section that the hydrocarbons sector

has much higher price elasticity than other sectors, while the

IAA sector has relatively high short-term demand elasticities

compared to other export sectors.




127

4. Possible Extensions of Econometric Analysis

4.1 A Foreign Trade Model (in volume and in price)

For each sector i, an import equation as well as export/import

price equation could supplement the modelling of Tunisia's

foreign trade.

4.1.1. Import Volume Function

The determinants usually adopted in the volume of imports are

domestic demand, a term in competitiveness formulated as the

relative price of domestic production compared to import prices

(usually calculated excluding energy) and a term in productive

capital utilization. Usually, the cyclical economic pressures on

production capacity are described by integrating this equation

into the utilization rates (UR) of domestic production capacity

relative to those of key partners. This ratio helps to identify a

possible supply constraint which is subject to the national

economy. The expected sign of its elasticity with respect to imports

is positive: when utilization rates are higher in Tunisia than in its

main partners, the increased domestic demand is directed towards

foreign producers, thereby increasing the volume of imports. If

this ratio comes out significantly in this relationship, its

consideration may, however, create some variational problems.

Lastly, some models enrich the analysis by incorporating non-

price competitiveness such as effort in research and development

(for example the integration of the age of capital).

The import function is written as follows:

Γi: the share of domestic products in domestic demand (household

consumption + investment by businesses) is a function of the ratio

between foreign and domestic prices (competitiveness effect).

4.1.2. Export Price Function

In fixing their prices (PX), Tunisian producers are alleged to have a

margin-driven attitude towards foreign and domestic markets alike.

Nevertheless, to cope with foreign competition, they also take

account of foreign prices (P*X) when setting export prices. Hence,

there is a trade-off between the margin-driven attitude (passing on

the total fluctuations in unit cost22 (CU) to export prices, so as to

maintain a constant profit margin), and a competitiveness-driven

attitude (passing on the total fluctuations in foreign prices to export

prices in a bid to maintain competitiveness). This trade-off translates

into a long-term target expressed as a weighted average of foreign

prices and domestic costs.

The import price function is written as follows:

19 An approximation of unit costs may be made by incorporating domestic production prices.


128


A value ε1 in the upper end means the competitiveness-driven

attitude prevails over the margin-driven attitude.

4.1.3. Import Price Function

Importers conduct a trade-off similar to that of exporters: in order

to maintain profit margins, they index their selling price on Tunisian

territory to their production costs, approximated by foreign import

prices (P*M). However, in order to maintain their competitiveness in

relation to French commodities, they also take into account domestic

production prices (PVA). Unlike foreign export prices, the foreign

import price is derived from simple weighting, given that competition

takes place only on Tunisian territory and therefore does not take

third-country markets into account.

The import price function is written as follows:

4.2 VAR Modelling from the Cointegration Equation

Based on the cointegration equation presented in section I:

miT(t) = a0 y(t) + a1 [piE(t) - py(t) ] + a2 [piT(t) - piE(t) ] + c + ê(t)

It is possible to proceed with the estimation of a VAR model, in order

to conduct forecast exercises. In this case, the structure of the VAR

model could be as follows (given that lag orders may be higher):

ΔmiT(t) =

b11ΔmiT(t-1) + b21Δy(t-1) + b31Δ [piE(t-1) - py(t-1)] + b41Δ [piT(t-1) -

piE(t-1)] + c01 – β1ê(t-1) + ε1(t)

Δy(t) =

b12ΔmiT(t-1) + b22Δy(t-1) + b32Δ [piE(t-1) - py(t-1)] + b42 Δ [piT(t-1) -

piE(t-1)] + c02 – β2ê(t-1) + ε2(t)

Δ [piE(t) - py(t)]

= b13ΔmiT(t-1) + b23Δy(t-1) + b33Δ [piE(t-1) - py(t-1)] + b43Δ [piT(t-1) -

piE(t-1)] + c03 – β3ê(t-1) + ε3(t)

Δ [piT(t) - piE(t)]

= b14ΔmiT(t-1) + b24Δy(t-1) + b34Δ [piE(t-1) - py(t-1)] + b44Δ [piT(t-1) -

piE(t-1)] + c04 – β4ê(t-1) + ε4(t)

However, it is worth noting that estimating such a model for each

sector is a huge task, and it is probably possible only after a selection

of the most important sectors for analysis (or considering only Tunisian

exports to Europe as a whole).

4.3 Non-linearities

The long-term structural equation for imports:

miT(t) = a0 y(t) + a1 [piE(t) - py(t) ] + a2 [piT(t) - piE(t) ] + c + ê(t)

hinges on the assumption of constant elasticities. However, various

forms of non-linearities or structural changes may be considered.

4.3.1. Temporal variation in income elasticity

For example, one may consider that income elasticity depends on

the European economy: in the early stages of the economic cycle

(for example, when unemployment rate u is higher than the natural

rate û), the income elasticity may be higher than in the low phases

of the cycle (when the unemployment rate is below the natural

rate).In order to model this process, a formalization based on

nonlinear smooth transition models (Smooth Transition) may be

proposed.

Suppose the following transition function, bound between 0 and 1,

wherein u(t) stands for the European unemployment rate, û the

natural unemployment rate and g>0 a parameter driving the velocity

of transition between regimes:

G(u(t) , g , û) =

It is established that if the unemployment rate is much higher than

129

the natural rate (on the brink when g.[u(t)-û] tends to infinity, which

is a mere mathematical assumption), function G tends towards

0, whereas when unemployment rate is much lower than the

natural rate (on the brink when g.[u(t)-û] tends towards minus

infinity, which is yet a mere mathematical assumption), function

G tends towards 1.

The proposed transition function helps to model a change in income

elasticity with respect to the European cycle of activity.

Indeed, the import equation may then be written as follows:

miT(t) = d0 y(t) + G.d1.yt + a1 [piE(t) - py(t)] + a2 [piT(t) - piE(t) ] + c + ê(t)

with: G =

Given that the value of G depends on the European unemployment

rate, the estimation of this equation will help to obtain income elasticity

values that necessarily range between d1+d0 (whenever the

unemployment rate is trending low) and d0 (whenever the

unemployment rate is trending high)23 .

The graph below represents a hypothetical simulation of income

elasticity engendered by such modelling wherein various values of

the unemployment rate are considered u(t) between 1% and 13%

and for the values of parameters fixed as follows:

g=50, û=6%, d0=0.1, d1 = 1.5.

At the econometric level, the estimation of the previous model may

be conducted through the non-linear least-squares method or the

least likelihood method, with a view to determining the values of

unknown d0, d1, a1, a2, c, g and û.

4.3.2. Temporal variation in price elasticity

The price elasticity of foreign trade may depend on the absolute

difference between Tunisian export prices and competitor export

prices piT(t) - piE(t).

Indeed, when the price differential is small, i.e. when [piT(t-1) -

piE(t-1)]² is close to zero (or a given threshold k), the price elasticity

of Tunisian exports may be assumed to be relatively low, whereas

when the price differential is huge, i.e. when [piT(t-1) - piE(t-1)]² departs

significantly from zero (or a threshold k), Tunisian exports will be

heavily dependent on fluctuations in relative prices.

In order to model such phenomenon, the following formalization

may be proposed:

Suppose the transition function G (.), bound between 0 and 1, where

in [piT(t-1) - piE(t-1)] stands for price differential, k the threshold beyond

which it is advantageous for consumers to change the content of

their consumption basket and h> 0 a parameter driving the velocity

of transition between regimes:

G([piT(t-1) - piE(t-1)] , h , k) =

23 Given that the unemployment rate ranges between 0 and 1, an alternative would consist in simply replacing function G with the unemployment rate observed. However,the proposed transition function has the advantage of taking into account broader transition variables than the unemployment rate, and which are not bound between 0and 1.


130


It is established that when the price differential is very high (positively

or negatively) with respect to the threshold k (on the brink as –h{[piT(t-

1) - piE(t-1)]² - k} tends to infinity), function G tends to 1, whereas when

the price differential remains low (in the sense that the distance between

the price remains close to the threshold k), function G tends to 0.

Hence, the proposed transition function makes it possible to model

a change in price elasticity based on the absolute difference in

relative prices.

Thus, the import equation can then be written as follows:

miT(t) = a0 y(t) + a1 [piE(t) - py(t) ] + f0 [piT(t) - piE(t) ] + G. f1 [piT(t) - piE(t)

]+ c + ê(t)

with: G =

Since the value of G depends on the absolute difference in relative

prices, the estimation of this equation will yield price elasticity values

between f1+f0 (to which it tends when the price differential is high)

and f0 (toward which it tends when the spread between the prices

is small).

The graph below represents a simulation of income elasticity

engendered by such modelling wherein various values of the price

differential are considered between -35% and 35% and for the

values of parameters fixed as follows: h=15, k=0, f0=-0.1, f1 = -1.5.

As in the previous case, the estimation of the model may be carried

out by the non-linear least squares method or maximum likelihood

method to determine the value of unknown parameters f0, f1, a0, a1,

c, h and k.

4.4 Quantitative Rationing by Supply or Demand

It has been stated in the foregoing that the Tunisian sector export

equation:

miT(t) = a0 y(t) + a1 [piE(t) - py(t) ] + a2 [piT(t) - piE(t) ] + c

actually describes a Europe-driven demand equation. From this

perspective, it may be relevant to define such demand equation by

stating it:

(4) DmiT(t) = a0 y(t) + a1 [piE(t) - py(t) ] + a2 [piT(t) - piE(t) ] + c

Similarly, the sector’s export supply (i) is conventionally modelled

as24:

SMiT(t) = γS YiT(t) [PiTX(t) / PiTX(t)]σT

Where:

SMiT(t): Tunisian export volume supply of commodity (i)

YiT: Tunisian total production volume of commodity (i)

PiTD: price index of commodity (i) on the Tunisian domestic market

(in local currency)

PiTX: Tunisian export price index of commodity (i), in local currency.

24 cf. Annabi and al., 2003.

131

γS: scale parameter

σT: verifying elasticity of processing σT>0

Suppose, in logarithms:

(5) SmiT(t) = yiT + σT [piTX(t) - piTX(t)] + c1

In the case of perfect price flexibility, the balance between supply

(7) and demand (6) will be achieved through an appropriate

adjustment of export prices25. However, if it is assumed that there

is some export price rigidity, the quantity exported will stand at least

between supply (6) and demand (7).

An achievable estimation of a quantitative rationing model of this

type may be conducted through a CES function as follows:

(6)

wherein the supply and demand functions are defined by

equations 6 and 7. Indeed, for large values of parameter ρ, the

CES function operates as operator Min.

The graph below illustrates the behaviour of the CES function with

respect to two time-varying variables S and D, where ρ = 100. It

can be observed that the CES function goes well with the minimum

of S and D.

Although the estimation of a CES function is not hitch-free, it may

be possible to econometrically estimate equation 8, in conjunction

with defining equations 6 and 7.


!

25 However, it is worth noting that PiTX differs from PiT, given that it does not factor in foreign exchange conversion, nor customs duties nor other costs borne by Europeanimporters of commodity i

132


This study sought to lay the groundwork for the modelling of

Tunisian sector exports towards the European market.

In the medium term, the completed study should help to carry

out medium-term forecasts (conditionally using international

environment scenarios) and provide decision makers with some

key guidelines to economic and industrial policies with a view to

enhancing the dynamics of Tunisia’s exports.

Indeed, quantitatively speaking, this analysis led to the identification

of areas most sensitive to changes in relative prices. From this

perspective, the orientation of specialization towards goods that

are less standardized, but more specific in terms of quality, should

reduce the sensitivity of certain exports to relative prices (this is

particularly true of the Textile, Apparel and Leather Sector and the

Chemical Industries Sector).

In the shorter term, these sectors would prove most sensitive to

deterioration in price competitiveness, which should guide policy

discussions aimed at improving their competitiveness. In addition,

the Hydrocarbons Sector, which turns out to be the most sensitive

to changes in relative prices, remains a very specific case due to

the standardized nature of its commodities.

With respect to elasticities of short-term demand, it was

demonstrated that the most sensitive sector is the Agri-food industry.

As a result, it is obviously most vulnerable to cycles in the global

economy. Nevertheless, it would be strategically profitable to

develop this sector's exports for the world's most dynamic

domestic markets.

Lastly, some modelling avenues were proposed as possible

extensions of the econometric analysis. At this stage, attempts to

estimate these models have proven to be problematic. According

to the RESET tests applied to various sectors, assumptions of non-

linearities were rejected, whereas the estimates of VAR models and

quantitative rationing models did not prove conclusive. Nevertheless,

these avenues deserve to be explored in future research works.

Conclusions

133

Bibliography

Annabi N., Cockburn J., Decaluwé B. (2003), Formes

Fonctionnelles et Paramétrisation dans les MCEG, CREFA,

Université de Laval.

De Boeff, S. (2000), Modeling Equilibrium Relationships: Error

Correction Models with Strongly Autoregressive Data, Political

Analysis, Vol 9, 14-48.

Dickey, D.A., and Fuller, W.A. (1981), Likelihood Ratio Statistics for

Autoregressive Time Series with a Unit Root, Econometrica, Vol 49,

pp 1057-72.

Engle, R.F., and Granger, C.W.J. (1987), 'Cointegration and error

correction: representation, estimation and testing, Econometrica,

Vol 55, pp 251-276.

Hurlin, C. (2001), L’Econométrie des Données de Panel, Ecole

Doctorale Edocif, Séminaire Méthodologique.

Narayan P.K. (2004), Reformulating Critical Values for the Bounds

F-statistics Approach to Cointegration: An Application to the Tourism

Demand Model for Fiji. Discussion Papers No. 02/04 Monash

University.

Pesaran, M.H., Shin, Y., and Smith, R.J. (2001), Bounds testing

approaches to the analysis of level relationships. Journal of Applied

Econometrics, Vol 16, pp 289-326.

Wong, K. N. (2008), Disaggregated export demand of Malaysia:

evidence from the electronics industry. Economics Bulletin, Vol.

6, No. 6 pp. 1-14.


Chapter III

Analysis of the Demand for Tunisian Goods

135


Table of Contents

136 Introduction

136 1. Stage One

141 2. Stage Two

145 3. Industry-by-Industry Analysis

136


Introduction

The purpose of the analysis is to identify "promising"

commodities for Tunisia. The methodology followed comprises

two major parts. In the first part, potentially "significant"

commodities for Tunisia are identified. The typology is based on

four main criteria: level of exports (global and vis-à-vis the European

Union) by Tunisia; level of revealed comparative advantage (RCA);

and variation in exports and revealed comparative advantage. For

variations (either of exports or RCA), the reference period is 2003-

2008, so as to better identify the dynamics involved. With respect

to levels, the calculation was done by taking the mean between

2006 and 2008, with a view to eliminating business cycle variations.

From the COMTRADE database at the most disaggregated level

(6-digit), and using the 4 criteria, 30 industries were identified,

which together account for 25% of Tunisia's exports in 2008.

While the first stage focuses on Tunisian supply, the second stage

lays emphasis on demand. Consequently, the changes in supply

for each of the 30 industries were analysed.

1. Stage one

The industry list given in Table 1 below provides relevant information

on the industries and other issues of interest. First, at the most

aggregate level, there are 7 agri-foodindustries (HS01 - HS23), 3

inorganic chemical industries (HS28; phosphates), and several W

"commodities" derived from iron and steel (HS72), and electrical

machinery (85). Almost all of these industries have a positive

trade balance and for most of them, the import level is very low.

They are mainly exporting industries, with very low intra-industry

trade. Regarding the level of exports, 4 industries account fort a

significant share (12%) of Tunisian exports: 150910, 280920,

310310, and 853690. Lastly, it is noticeable that all the industries

selected have a positive and very high revealed comparative

advantage, except for 852812.

In order to better appraise the significance and trends of these

30 industries, they were sorted by level of importance among

Tunisian Exports and then divided into three groups. The following

pages contain graphs, each for one of the three groups. For all

industries, the graphs are for the period 2000-2009. There are

three graphs for each industry group: one shows the evolution of

exports (and then it also gives the opportunity to see the sharp

drop in trade due to the financial crisis), the second shows the

change between 2003 and 2008 and the third shows the trend

of the revealed comparative advantage.

137


Product Product Name RCA Exports Trade Balance Exp.Share

030239 Tunas skipjack or stripe-bellied bonito... 0.96 58,382.56 58,381.12 0.30%

040630 Processed cheese, 0.81 29,775.37 28,811.48 0.15%

150910 Virgin 0.98 574,217.43 572,761.60 2.97%

150990 Other 0.93 46,095.95 46,053.40 0.24%

151000 Other oils..... 0.97 27,738.70 23,783.47 0.14%

151710 Margarine, excluding liquid margarine 0.87 38,352.88 38,352.31 0.20%

230690 Other 0.93 9,216.26 9,216.26 0.05%

251010 Unground 0.94 147,511.96 147,511.96 0.76%

280920 Phosphoric acid and polyphosphoric acids 0.97 725,131.54 639,897.35 3.75%

283525 Phosphates: Calcium hydrogenorthophosphate 0.98 64,320.84 64,298.90 0.33%

283526 Phosphates:-- Other phosphates of calcium 0.97 90,545.08 90,523.38 0.47%

310310 Superphosphates 0.99 626,892.66 626,892.66 3.25%

520839 Dyed :-- Other fabrics 0.76 11,031.69 -40,415.97 0.06%

611249 Women's or girls' swimwear 0.99 35,060.86 33,570.27 0.18%

621010 Of fabrics of heading No. 56.02 or 56.03 0.97 158,879.24 146,244.53 0.82%

721030 Electrolytically plated or coated with zinc 0.52 22,776.00 19,676.60 0.12%

721049 Otherwise plated or coated with zinc 0.59 116,334.99 91,388.48 0.60%

721491 Other :-- Of rectangular cross-section 0.65 14,617.94 5,840.73 0.08%

740620 Powders of lamellar structure; flakes 0.97 20,300.78 20,244.92 0.11%

847190 Other 0.80 81,425.79 72,110.28 0.42%

851750 Other apparatus, for carrier-current line systems or for digit... 0.80 161,590.92 139,091.27 0.84%

852812 Reception apparatus for television 0.23 170,515.88 165,936.32 0.88%

853180 Other apparatus 0.93 68,192.05 63,597.62 0.35%

853690 Other apparatus 0.87 578,730.27 191,882.98 3.00%

854430 Ignition wiring sets and other wiring sets 0.70 172,225.15 -10,089.72 0.89%

854459 Other electric conductors 0.76 239,581.68 192,839.69 1.24%

854890 Other 0.87 89,169.01 69,144.95 0.46%

902830 Electricity meters 0.89 36,870.47 36,411.57 0.19%

961210 Ribbons 0.84 25,165.10 14,132.50 0.13%

961390 Parts 0.98 22,963.88 20,068.90 0.12%

Total ######### 3,578,159.80 0.23%

Table 1: Key Industries in Tunisia

138


Graph 1.a: Evolution of exports of the key industriesTop 10 industries

Graph 1.b: Tunisia Exports to World 2003, 2008Top 10

Graph 1.c: Tunisia RCA with World 2000-2009Top 10

139


Graph 2.a: Evolution of exports of the key industriesTop 11-20

Graph 2.b: Tunisia Exports to World 2003, 2008Top 11-20

Graph 2.c: Tunisia RCA with World 2000-2009Top 11-20

140


Graph 3.a: Evolution of exports of the key industriesTop 21-30

Graph 3.b: Tunisia Exports to World 2003, 2008Top 21-30

Graph 3.c: Tunisia RCA with World 2000-2009Top 21-30

141


2. Stage Two

After identifying potentially successful industries with respect to

Tunisian supply, the second stage attempts to assess the demand

trend and compare it with Tunisian supply.

First, there is a change in global demand for the 30 industries

identified, as shown in the graph below:

Graph 4: World Imports from World 2003, 2008

The graph describes global imports for these products for two years,

2003 and 2008, and the relative level of, and variation in, the demand

for each commodity. Five industries have a global demand much higher

than the rest: (721049, 852812, 853690, 854430, and 8545459).

Industries for which demand has increased the most are 251010, 230690,

310310, 854459, and 280920. For each of these five industries, global

demand has increased by more than 300%. During this period, the

increase in total global imports was a little over 110%. Twelve industries

had higher demand and two industries (030239, 851750) experienced

a decline in global demand (34% and 43%). See also table below.

142


Reporter Product Product Name 2003 2008 2008 % Variation

0 030239Tunas (of the genus Tunnus) skipjack or stripe-bellied bonito (Euthynnus (Katsuwonus) pelamis),exc lunding livers and roes : -Other

556,830.991 366,877.877 -0,341

0 040630 Processed cheese, not grated or powdred 1,310,064.957 2,383,704.616 0.820

0 150910 Virgin 2,534,737.640 4,907,778.465 0.936

0 150990 Other 788,101.796 1,256,834.297 0.595

0 151000

Other oils and theirfractions, obtained solely fromolives, wether or not refined, but not chemiccalymodified, including blends of these oils or fractionswith oils or fractions of heading N°.15.0

85,354.027 230,863.345 1.705

0 151710 Margarine, excluding liquid margarine 742,651.639 1,624,874.252 1.188

0 230690 Other 65,728.005 303,258.366 3.614

0 251010 Unground 741,393.439 3,792,910.516 4.116

0 280920 Phosphoric acid and polyphosphoric adds 1,760,366.957 7,202,665.315 3.092

0 283525Phosphates: -calcium hydrogenorthophosphate(“dicalcium phosphate”)

264,012.468 633,375.252 1.399

0 283526 Phosphate:-other phosphates of calcium 297,946.680 1,059,238.644 2.555

0 310310 Superphosphates 627,845.727 2,700,043.471 3.300

0 520839 Dyed:-other fabrics 682,166.423 742,462.914 0.088

0 611249Women’s or girls’ swimwear:-of other textilematerials

63,634.259 64,613.937 0.015

0 621010 Of fabrics of heading N°. 56.02 or 56.03 951,616.144 1,488,329.897 0.564

0 721030 Electrolytically plated or coated with zinc 3,541,817.987 7,085,701.724 1.001

0 721049 Otherwise plated or coated with zinc:-other 9,291,254.117 22,466,215.247 1.418

0 721491Other:-of rectangular (other than square) cross-section

751,400.926 2,279,521.149 2.034

0 740620 Powders of lamellar structure; flakes 112,428.911 132,925.421 0.182

0 847190 Other 4,831,012.368 6,967,781.969 0.442

0 851750Other apparatus, for carrier-current line systemsor for digital line systems

17,359,352.886 9,886,454.734 -0.430

0 852812

Reception apparatus, for teleevision, whether ornot incorporating radio-broad cast receivers orsound or video recording or reproducing appa-ratus:-colour

26,404,762.434 78,694,420.177 1.980

0 853180 Other apparatus 2,038,031.989 2,282,609.436 0.120

0 853690 Other apparatus 16,462,291.070 31,350,854.230 0.904

0 854430Ingnition wiring sets and other wiring sets of akind used in vehicules, aircraft or ships

14,839,577.745 23,516,802.519 0.585

0 854459Other electric conductors, fora voltage exceeding80v but not exceeding 1,000 v:-other

5,282,229.260 22,222,484.002 3.207

0 854890 Other 2,761,272.151 3,242,141.436 0.174

0 902830 Electricity meters 822,344.712 1,658,544.995 1.017

0 961210 Ribbons 1,345,815.367 1,706,126.056 0.268

0 961390 Parts 101,217.088 142,548.323 0.408

Table 2: World trade with world 2003,2008

143


Second, it is interesting to see which exporting countries are

(potentially) competitive with Tunisia. To achieve this, the industries

should be considered individually and collectively to identify countries

that are strong competitors in a number of key sectors. For the

industry-by-industry analysis, see section below.

To isolate the major competitors in the 30 sectors, the ten largest

exporters in the world were first identified, for 2008, and for each of

the sectors, and the number of appearances of each country as the

largest exporter or importer in the selected industries was counted.

Then, the following table shows all countries that emerge from this

analysis and appear at least five times. The first column gives the

number of times each country appears as one of the ten most

significant exporting countries and the second column shows the

number of times each country appears as one of the most significant

importing countries. Hence, the first line reveals that among the 30

countries identified, China is one of the major exporters (competitor?)

21 times, followed by Germany, France and the United States. It is

interesting that Tunisia stands out as one of the 10 most significant

countries for half of the industries - i.e. 15 times.

After identifying the "competitor" countries, it is interesting to assess

the level of similarity in the structure of exports between these

countries and Tunisia. The revealing indicator is Finger-Kreinin (FK).

It allows a comparison between the export structures of two

countries. If the structures are identical, FK is equal to "1", in case

both countries have totally different structures, which means there

is no commodity exported by both countries, FK is equal to "0".

Table 4 provides the FK between Tunisia and all countries identified

in the table above. It appears the level of similarity is highest with

Morocco and Mexico (0.346 and 0.334), followed by Turkey, and

then some European countries. The level of similarity is lowest with

Israel, Japan, Korea, and Taiwan. An indicator value of 0.346 can

be interpreted as a degree of similarity of export structures of

approximately 34.6%. As a benchmark, the level of similarity

between the U.S. and the EU is typically about 65%.

These statistics are very interesting. The level of similarity with all

countries (except Morocco and Turkey) increased. FK with

Morocco declined slightly over this period, indicating a

differentiation in specialization by both countries in different

commodities, but with almost no change recorded for Turkey.

The increase in similarity is most significant with Japan, Mexico,

and Israel. Given the level of change and similarity, it seems the

biggest competitors globally are some European countries (the

Czech Republic, France, Italy, Spain), and Morocco (but with a

decrease over the years).

The FK index considers the level of similarity between two countries

at the level of their exports structure. The RECPI index takes into

account the structure and level of exports. Supposing there are

two very similar countries, two cases are possible. In the first case,

the countries have nearly the same size, and in the second case,

there is a huge difference in size. FK does not measure or does

not capture this difference. This is done by RECPI which will then

report that the competitive pressure is likely to be much higher in

the second case as compared to the first. Hence, the table below

gives the RECPI for Tunisia and 18 countries by considering their

global exports. The greater the figure, the stronger the competitive

pressure.

Table 3: 2008 - Number of times each country appears asa principal exporter or importer

Country Export Count Import Count

CHN 21 8

DEU 18 21

FRA 16 23

TUN 15 2

USA 15 21

ITA 13 16

ESP 12 18

NLD 12 14

BEL 11 12

GBR 10 20

JPN 10

MEX 9 8

TUR 9 1

ISR 6 1

KOR 6 5

CZE 5 1

MAR 5 1

POL 5 7

TWN 5 3

144


Table 4: FK Index between Tunisia and selected countries on exports to the world

Country 2003 2004 2005 2006 2007 2008

BEL 0.160 0.137 0.157 0.173 0.177 0.185

CHI 0.205 0.187 0.199 0.206 0.211 0.210

CZE 0.153 0.154 0.178 0.176 0.184 0.191

FRA 0.165 0.161 0.183 0.194 0.203 0.204

DEU 0.152 0.147 0.167 0.177 0.180 0.181

ISR 0.080 0.078 0.081 0.087 0.099 0.110

ITA 0.205 0.199 0.222 0.232 0.232 0.236

JPN 0.092 0.092 0.105 0.117 0.125 0.132

KOR 0.126 0.104 0.117 0.131 0.134 0.141

MEX 0.245 0.261 0.302 0.293 0.333 0.334

MAR 0.380 0.384 0.376 0.353 0.342 0.346

NLD 0.164 0.141 0.157 0.166 0.167 0.182

POL 0.174 0.166 0.192 0.198 0.200 0.202

ESP 0.192 0.188 0.206 0.214 0.217 0.231

TUR 0.268 0.250 0.267 0.270 0.267 0.265

GBR 0.207 0.194 0.212 0.225 0.238 0.251

USA 0.144 0.140 0.155 0.171 0.167 0.179

TWN 0.122 0.113 0.125 0.145 0.143 0.149

Table 5: Indice RECPI de la Tunisie et d’un groupe de pays concurrents

Country 2003 2004 2005 2006 2007 2008

BEL 3.36 1.99 2.48 3.18 2.19 2.55

CHI 4.95 3.65 4.82 5.62 4.08 3.99

CZE 0.33 0.27 0.37 0.39 0.30 0.35

FRA 2.52 1.56 2.04 2.55 1.56 1.72

DEU 4.04 3.10 3.82 4.52 2.95 2.80

ISR 0.12 0.07 0.06 0.07 0.06 0.12

ITA 3.39 2.31 2.77 3.32 2.33 2.22

JPN 1.44 1.21 1.49 1.81 1.36 1.68

KOR 2.08 1.30 2.08 2.86 1.93 2.59

MEX 9.07 9.30 13.44 14.73 13.01 11.04

MAR 0.48 0.39 0.36 0.33 0.22 0.41

NLD 4.02 2.10 3.02 4.20 2.87 3.72

POL 0.47 0.35 0.46 0.55 0.40 0.44

ESP 1.75 1.48 1.52 1.95 1.20 1.37

TUR 1.16 0.88 0.90 0.97 0.72 0.75

GBR 8.83 7.79 10.26 11.31 9.52 8.71

USA 3.94 2.45 3.27 4.64 3.01 4.78

TWN 1.24 0.87 1.28 1.70 1.12 1.23

The table reveals (generally) that the most competitive country is

Mexico, followed by the United Kingdom, the United States, China

and Nederland. It is interesting to note that the countries where

he competitive level increased the most are Korea, Japan and the

United States. In contrast, the competitive level decreased with

respect to most of the other countries.

145


Table 6: Concurrents de la Tunisie par secteur

Country Mineral Fuels Electrical Machinery Apparel & Clothing Fertilisers Inorganic Chemicals

27 85 62 31 28

CZE 0.217 0.350 0.587 0.005 0.105

TUR 0.212 0.322 0.521 0.517 0.143

GBR 0.712 0.247 0.444 0.155 0.023

USA 0.244 0.231 0.468 0.575 0.070

BEL 0.254 0.265 0.513 0.077 0.172

CHI 0.306 0.237 0.461 0.307 0.134

FRA 0.214 0.342 0.434 0.032 0.016

DEU 0.214 0.284 0.504 0.010 0.024

ISR 0.217 0.139 0.283 0.219 0.300

ITA 0.246 0.316 0.442 0.078 0.033

JPN 0.214 0.216 0.426 0.006 0.023

KOR 0.213 0.167 0.259 0.226 0.038

MEX 0.911 0.298 0.501 0.750 0.183

MAR 0.212 0.360 0.509 0.742 0.660

NLD 0.218 0.221 0.540 0.041 0.060

POL 0.215 0.358 0.508 0.025 0.112

ESP 0.214 0.330 0.446 0.092 0.117

TWN 0.213 0.145 0.294 0.001 0.030

3. Industry-by-Industry Analysis

One of the industries with the strongest growth was 854459 (other

electrical conductors). Under the period 2006-08, this industry

was the 25th industry exported by Tunisia in 2006 and 13th in

2008, with an export share which increased from 0.81% to 1.24%

and an export growth rate slightly above 150%. The standardized

relative comparative advantage index also increased significantly

from 0.51 to 0.76, starting 2003.

The table below shows the 20 largest exporters of the world (and then

potentially the most significant competitors, as well as the 20 largest

importers of the world (which shows where the highest demand for

this industry is), and finally the 20 most important destinations for that

industry for Tunisia. The table also gives the rank for each of the latter,

including the percentage change in trade between 2006 and 2008.

There are several interesting aspects in this table. First, it is noticeable

that 14 of the largest exporters are also major importers. This may

The table below reviews the level of similarity, by sector, for the five

largest Tunisian exports sectors at 2-digit HS level. They account

for over 60% of exports. For each of these sectors, the comparison

involves the level of similarity between Tunisia and the 18 countries

identified as key competitors. In the table, for each sector, the five

countries with the highest level of similarity are shown in red, and

the most significant in "bold". There are noticeable differences among

the sectors regarding the most significant competitors. Morocco is

one of the 5 most significant competitors in four sectors; and 3

countries, namely: Turkey, Belgium, and China in 3 areas. However,

the variance is quite high across sectors. If "Electrical Machinery" is

taken into consideration, many countries will have fairly identical

levels of similarity. However, as regards the inorganic chemicals

sector, Morocco stands out as probably the biggest competitor.

Consequently, it is important to consider the policies in these countries

and the development of their trade in key sectors when formulating

subsequent relevant policies. To achieve this, a sector-by-sector

analysis is rather recommended, as shown in the section below.

146

suggest that this is an industry/commodity where vertical integration

in "supply chains" is significant or an industry where, at the 6-digit

HS level, there is great horizontal differentiation. In the first case,

the high intersection between exporting and importing countries

suggests their involvement in the "value added chain." In the second

case, it suggests that it is an industry with highly differentiated

commodities, even at a highly disaggregated level.

Second, it is observed that the countries that posted the highest export

growth (above 200%) are Japan, Hungary, Austria, China, Turkey, the

United States and Norway. Those that recorded the highest import

growth are Romania, Mexico, China, Poland, Czech Republic, Slovakia

and Holland. Third, it is very interesting to see the intersection between

the main importers of the world and Tunisian export destination. Here,

there is an intersection with only six countries (Germany, Italy, France,

Austria, UK, and Belgium), all of which are European Union members.

This could perhaps suggest that this is an industry where Tunisia enjoys

good and increasing competitiveness, but fails to access the many

other markets. This could be for good reasons such as distance,

differences in demand, lack of knowledge of these markets on the part

of Tunisian exporters, etc., and may as well indicate the possibility of

barriers in other markets.


World Tunisia

CZE 1 180 1 153 2 20339USA 2 217 2 145ITA 3 144 11 172 16 -79CHN 4 297 4 266POL 5 139 13 226FRA 6 169 5 150 1 221TUR 7 234JPN 8 262 12 158MEX 9 133 6 301CZE 10 176 14 237ESP 11 194AUT 12 284 8 135 11 NewGBR 13 125 3 133 9 -84KOR 14 197SWE 15 144BEL 16 173 17 196 10 354HUN 17 357CAN 18 109 7 172NLD 19 171 15 206NOR 20 204HKG 9 151ROM 10 479HUN 16 212SVK 18 258THA 19 144SGP 20 160GNQ 17 NewCOM 18 NewCMR 19 3019EGY 20 NewETH 3 441LBY 4 925DZA 5 178BFA 6 20IRQ 7 1373COG 8 32MLI 12 -18MDG 13 -48SEN 14 2006 onlyMAR 15 -87

Table 7: Exportations mondiales et tunisiennes vers un groupe de pays

147


The table below shows Tunisian exports during the 2000s towards

the two most important destination countries for Tunisia (Germany

and France) and among the fifteen countries that are the world's

most significant importing countries, the eight countries that do not

appear as an important destination for Tunisia. It is noticed that

France is an important, long-standing destination for Tunisia, with

fairly sharp increase until 2008. By contrast, Germany has become

a popular destination in recent years. Furthermore, it would be very

interesting to look at Tunisia's sector exports compared to other

countries. It is revealed that for most of the years, and with respect

to most countries, Tunisia almost does not export, except for a very

low value in 2008.The latter may be random or may perhaps indicate

the export trend of this commodity towards a wider range of

countries.

Table 8: HS 854459: Tunisian Exports

Partner 2000 2001 2002 2003 2004 2005 2006 2007 2008

PRA 1,888.12 1,608.78 1,577.27 483.97 693.92 4,736.90 22,574.91 16,204.95 72,451.40

DEU 283.05 23.07 151.74 0.64 0.01 0.25 138.05 349.69 28,215.38

CAN 0.02 0.00 0.22 0.00

CHN 129.25 0.00 0.06 0.00 0.00

CZE 0.00 0.00 0.00 129.52

HKG 1.79

JPN 0.00 0.00 0.00 4.39

MEX 0.00 87.52

ROM 5.02

USA 47.20 0.00 0.00 0.00 127.93

Chapitre IV

Assistance for the Analysis of Tunisia’s Production System

149


Table des matières

150 1. Summary of Work Done

150 1.1 Operation of individual databases

150 1.2 Determining a Methodological Framework for Analysing the Productivity Trend of the Tunisian Industry

152 2. Analysis of Outputs

152 2.1 Descriptive Analysis of Tunisia’s Industrial Business Database

164 2.2 Productivity Decomposition Analysis

164 2.2.1 Principle of Decomposition Methods

165 2.2.2 Outputs

171 3. conclusion

172 Bibliography

150


1. Summary of Work Done

Following the first mission fielded in September, the emerging

needs of ITCEQ focused primarily on two points: (i) the operation

of individual databases; and (ii) the determination of a framework

analysis to study the labour productivity trend of Tunisian businesses.

1.1 Operation of individual databases

One of the major concerns of the ITCEQ team concerning the use

of individual databases is the problem of cleaning. This part of the

work is crucial, given that it determines the quality of outcomes and

the analysis that will be performed from these data. Moreover, it is

important to follow a rigorous procedure and control successive

changes in the base that ensue from the various stages of cleaning.

This procedure, which is programmed on Stata, requires time and

proper mastery of the software.

The procedure involves two major steps. The first is to prepare the

base on Stata (creation of a single base from various data files,

harmonization of questionnaires, interpolations, etc.). The second

step is "cleaning properly speaking” i.e., to remove outliers and

observations when key variables are not populated.

Technical assistance provided by DEFI team consisted in:

- First, presenting the methods for cleaning the databases of

businesses most commonly used in the literature (see slides in

Appendix);

- Second, presenting and explaining the main commands for

managing databases on Stata (see slides in Appendix); and

- Third, creating, under Stata, a single database containing all the

available data from surveys which were originally on separate

Excel files by year and by category of variables (status, outcome,

employment, liabilities, assets, capital assets, identification,

accounting values, book values and suite).

The transition from these Excel files to a full database on Stata would

require the:

- harmonization of all Excel files to make them comparable;

- establishment of correspondences between different

questionnaires (questionnaire change from 98);

- Addition of price indices; and

- merger of all these files so as to have a complete STATA

package..

- Fourth, developing, with the ITCEQ team, the cleaning procedure

programme in which many comments have been introduced so

that ITCEQ can not only use it but also modify it to their liking.

Key outputs of part one (I.1)

- ITCEQ executives trained on methods of cleaning

microeconomic data and on the management of databases on

Stata

- In terms of output, ITCEQ has (i) a comprehensive database

within Stata with all the variables available from the Survey over

a period of 11 years (1997-2007); and (ii) a cleaning programme

on Stata that was validated in unison during the last mission

in April.

I.2 Determining a Methodological Framework forAnalysing the Productivity Trend of the Tunisian Industry

Discussions during the first mission in Tunis led to the decision to

apply the productivity decomposition method. This method consists

151

in identifying the extent to which productivity gains are attributable

to productivity growth within businesses or to the phenomenon of

reallocation.

Technical assistance provided by DEFI team consisted in:

- first, presenting and explaining the various productivity

decomposition methods used in the literature on individual data

(see slides in Appendix);

- Second, developing the selected decomposition programme on

Stata and explaining its content and the various stages of this

programme to the ITCEQ team, in order that its executives may

use it without depending on the DEFI team.

Key outputs of part two (I.2)

- ITCEQ executives were trained on the decomposition methods

used in the literature.

- In terms of output, ITCEQ has a productivity decomposition

programme under Stata.


152


2. Analysis of Outputs

In this first section, a descriptive analysis is conducted on thedatabase of Tunisian businesses from the sample of surveys over

a period ranging from 1997 to 2007. In the next section, the

outcomes from the productivity decomposition are analysed.

2.1 Descriptive Analysis of Tunisia’s IndustrialBusiness Database

All data in the database are derived from annual surveys conducted

by the National Institute of Statistics of Tunisia and made available

to ITCEQ. The database covers sectors of the Tunisian industry from

1997 to 2007. It contains information on production, intermediate

consumption, permanent employment, seasonal employment,

industry sector, the region and capital structure. The transition to

constant prices was carried out using the price indices for production,

value added and 5-digit price indices for intermediate consumption

provided by INS. The business performance indicator used is labour

productivity, obtained from each company, via the ratio of value

added at constant prices over the entire workforce, which comprises

both the permanent workforce and seasonal jobs.

By retaining only the industrial sector, the "raw" initial database has

16 442 remarks, representing 4,464 businesses. Once cleaned

(detailed cleaning procedure in Box 1), the unbalanced panel

database includes 15 202 remarks and 4206 businesses. Table 1

gives the number of businesses per year. It is record low in 2007,

with 1,180 businesses and record high in 2000, with 1,613

businesses.

Table 1: Number of Businesses per Year

Years Number of Businesses

1997 1380

1998 1471

1999 1385

2000 1613

2001 1586

2002 1333

2003 1253

2004 1247

2005 1318

2006 1436

2007 1180

Number of Remarks 15 202

153

Table 2 shows the number of businesses by number of years of

presence in the database. Note that this number of years of presence

may not be consecutive. It is found that only 89 firms are present

for all the years, i.e. 11 years running. In addition, a very large number

of enterprises (1,469 i.e. about 35% of businesses) appear only for

one year.


Table 2: Number of Businesses According to the Number of Years of Presence in the Database

Number of years of presence in the database Number of Businesses In percentage

1 1469 35%

2 561 13%

3 469 11%

4 327 8%

5 322 8%

6 318 8%

7 205 5%

8 178 4%

9 140 3%

10 128 3%

11 89 2%

Total 4206 100%

Table 3: Number of Businesses by Sector

Sector Number of Businesses In percentage

1 Agri-food and Tobacco 561 13%

2 Textile 262 6%

3 Apparel 1236 29%

4 Footwear and Leather 250 6%

5 Timber, Paper and Publishing 283 7%

6 Chemicals and Pharmaceuticals 179 4%

7 Rubber and Plastics 159 4%

8 Non-metal Materials 314 7%

9 Metal Materials 320 8%

10 Equipment, Machines and Electrical Appliances 329 8%

11 Automotive Industry and other Transportation Equipment 102 2%

12 Furniture 211 5%

Table 3 below shows the breakdown of the sample firms by sector.

Most prominently represented are Apparel (29%) and Agri-food (13%)

On their own, they account for 42% of businesses in the sample.

However, the Automotive sector accounts for only 2% of the number

of firms, followed by the Chemicals & Pharmaceuticals, Rubber &

Plastics sectors (4% each).

154


The breakdown of 4,206 businesses by size was conducted based

on the criterion of average total employment of each business by

using the quantiles method. The resultant breakdown classifies

companies with a total number of employees not exceeding 23

under the "small" group. In other words, the first third of the sample

firms have, on average, a number of employees not exceeding 23.

In the "average" group (which corresponds to the second third of

businesses), businesses have a number of employees strictly greater

than 23 and not exceeding 77. "Large" group (the last third of sample)

firms have more than 77 employees.

Table 4: Number of Businesses by Size According to the Number of Years of Presence in the Database


Number Percentage


1 1469 732 455 282 50% 31% 19%

2 561 195 205 161 35% 37% 29%

3 469 144 172 153 31% 37% 33%

4 327 84 104 139 26% 32% 43%

5 322 114 106 102 35% 33% 32%

6 318 68 109 141 21% 34% 44%

7 205 27 78 100 13% 38% 49%

8 178 17 67 94 10% 38% 53%

9 140 10 46 84 7% 33% 60%

10 128 3 24 101 2% 19% 79%

11 89 1 15 73 1% 17% 82%

Table 4 shows the breakdown of businesses by size and by the number

of years of presence in the database. The first row of the table shows,

for instance, that among the 1,469 firms present in a single year, half

of them (i.e. 732) fall under the "small" category, about a third (or 455)

belong to the "average" category and 19% (i.e. 282 companies) are

considered "large". Therefore, arrival and disappearance from the

sample (which, it should be recalled, are not necessarily new businesses

or cessation of activities) concern more of small firms. The distribution

by size of companies present for five years generally corresponds to

breakdown by quantile of 4206 businesses in the sample.

Approximately 80% of companies in the database for 11 or 10 years

fall under the "large" business category. On average, companies with

over 77 employees are present in the sample during a number of years

higher than the "average" and especially "small" categories.

Box 1. Cleaning Procedure

The procedure developed was largely inspired by Hall and Mairesse (1995).This eliminates firms that have never filled their VA or employment. In addition, variables with zero, unfilled or negative observations were removed, as well as those with annual growth ratesratios, VA/Total Employment, Intermediate consumption/Total Employment, Income/Total employment or Capital/Total employmentratios greater than 500 % or less than -500%.

155

Table 5 dwells on the breakdown by size of businesses in the

database within the various sectors. For example, the second part

of the table shows that the breakdown in sector 10 (Electrical

Machinery) is closest to the distribution of firms across the database.

30% of companies are indeed small-sized, 36% are medium-sized

and 35% are large. However, the figures in bold or underscored in

gray highlight where the percentages substantially differ from those

that correspond to all the sectors. The highest fall in bold and the

lowest percentages are highlighted in gray. It is found that companies

in the "small" category are relatively more present in the timber, paper

& publishing sector (58%), agri-food (57%), furniture manufacturing

(51%) and metal materials (45%) sectors. However, they are less

present in two sectors: apparel (10%) and footwear and leather

(23%). It is in this leather and footwear sector alone that "medium-

sized" companies are relatively the most active, with a 40% share.

Furthermore, it is only in one sector (food and agricultural) that these

"medium-sized" businesses are relatively less prevalent (21%). Lastly,

companies considered "large" are more strongly represented in

apparel (57%), but relatively less present in timber, paper & publishing

(13%), metal materials (17%) food and agriculture (22%).



SectorsNumber Percentage


1. Agri-food and Tobacco 561 317 119 125 57% 21% 22%

2. Textile 262 96 103 63 37% 39% 24%

3. Apparel 1236 124 406 706 10% 33% 57%

4. Footwear and Leather 250 57 101 92 23% 40% 37%

5. Timber, Paper & Publishing 283 163 82 38 58% 29% 13%

6. Chemicals and Pharmaceuticals 179 68 62 49 38% 35% 27%

7. Rubber and Plastics 159 61 62 36 38% 39% 23%

8. Non-metal Materials 314 124 109 81 39% 35% 26%

9. Metal Materials 320 145 121 54 45% 38% 17%

10. Electrical Machinery 329 98 117 114 30% 36% 35%

11. Automotive Industry 102 35 28 39 34% 27% 38%

12. Furniture 211 107 71 33 51% 34% 16%

Given that, as already highlighted, "large" businesses stay much longer

in the database than the "medium" and especially "small" categories,

the distribution of businesses by year and by size described in Table 6

shows first the prevalence of large enterprises in the total number of

observations. Indeed, they account for 46% of 15,202 observations in

the entire sample (6,996 observations concerning the category of "large"

businesses), the medium category accounts for 32% (4,882 observations),

while "small" firms make up only 22% of total observations (i.e., 3,324

observations). It is observed that in 2002, 2004 and 2005, "small" firms

are relatively poorly represented (with shares of 16%, 14% and 15%

respectively) mainly to the advantage of "big" companies, especially in

2004 and 2005, given that this category of "large" firms account for

56% and 55% of annual observations for both years, respectively.

Table 7 below shows the breakdown of firms by capital structure

and by size. Among the 4,206 businesses in the sample, 126 (3%)

have part of their capital held by the State and 1,243 (30%) have

part of their capital held by foreign investors. The firms concerned

fall mainly under the "large" category (58% for capital held by the

State and 65% for capital held by a foreign entity).

156


Table 8 shows the number of businesses by major region. It is

found that the vast majority of businesses in the sample are located

in the district of Tunis, the North East and Central East. Only

6% of the 4,206 firms are located in the North West, 3% are

located in the South (East and West) and 2% of the sample in the

Centre West.

Table 9 presents the averages and standard deviations of value

added, intermediate consumption, production, number of permanent

employees, total number of employees, capital stock, investment

and labour productivity, both for all businesses in the sample and

by breaking them down by size. The case of Tunisian businesses

confirms what is usually found in the literature, namely that productivity

increases with company size. Indeed, "large" companies have a

higher unweighted average labour productivity (8.78) compared

to "medium" firms (8.71) and "small firms" (8.48).

The value added, intermediate consumption, production and capital

stock of "medium” businesses are 4 times higher than those of

“small” businesses. Instead, they invest 5 times more and employ

only 3.5 times more than "small" firms. The "large" companies have

value added 24 times higher than smaller ones. They invest 28 times

Table 6: Number of Businesses by Year and by Size


Number Percentage


1997 1380 386 457 537 28% 33% 39%

1998 1471 417 478 576 28% 32% 39%

1999 1385 348 467 570 25% 34% 41%

2000 1613 368 586 659 23% 36% 41%

2001 1586 323 568 695 20% 36% 44%

2002 1333 219 449 665 16% 34% 50%

2003 1253 251 361 641 20% 29% 51%

2004 1247 174 380 693 14% 30% 56%

2005 1318 204 387 727 15% 29% 55%

2006 1436 336 417 683 23% 29% 48%

2007 1180 298 332 550 25% 28% 47%

Total (number ofobservations)

15202 3324 4882 6996

Total (in % of No. of obs.)

100% 22% 32% 46%

Table 7: Number of Businesses by Capital Structure and by Size

Corporate capitalstructure

Number Percentage by business type


Businesses thathave at least part oftheir capital held bythe State

126(i.e. 3% of 4206 businesses)

24 29 73 19% 23% 58%

investors (businesses) 107 334 802 9% 27% 65%

157

more, and hold capital stock 31 times higher, and their total

employment is 17 times higher, than the category of "small" firms.

Consequently, they are characterized by higher capital intensity

(almost 2 times higher than small, and a little over 1.5 times higher

than medium, businesses).This finding may suggest that by using

total factor productivity (TFP) in lieu of labour productivity, the

category of "large" companies may not have the highest average

level of efficiency.


Table 8: Number of Businesses by Major Region

Regions Number of Businesses In percentage

1 District of Tunis and North East 1829 45%

2 North West 246 6%

3 Centre East 1731 43%

4 Centre West 95 2%

5 South East and West 122 3%

6 Total 4023* 100%

* 183 businesses did not provide information on their location. Hence, the figures available fall short of the total number of businesses in the database, which stands at 4206.

Table 9: Key Statistics for Businesses in the Entire Sample and by Size

By Size Variables Mean Standard Deviation

Small

Value added (constant price) 101 583 303 478

Intermediate consumption (constant price) 329 622 1 438 669

Production (constant price) 428 824 1 681 111

Number of permanent employees 13 6

Total number of employees (permanent and seasonal) 13 6

Capital Stock (constant price) 165 439 282 812

Investment (constant price) 23 277 72 482

Labour Productivity * 8,48 0,90

Medium

Value added (constant price) 427 958 880 480Intermediate consumption (constant price) 1 246 976 2 953 823Production (constant price) 1 670 705 3 670 848Number of permanent employees 43 20Total number of employees (permanent and seasonal) 47 20Capital Stock (constant price) 662 209 1 067 453Investment (constant price) 116 032 328 370Labour Productivity * 8,71 0,90

Large

Value added (constant price) 2 496 077 8 067 454Intermediate consumption (constant price) 6 327 027 22 900 000Production (constant price) 8 807 453 30 100 000Number of permanent employees 205 261Total number of employees (permanent and seasonal) 232 334Capital Stock (constant price) 5 129 195 22 700 000Investment (constant price) 659 287 2 149 993Labour Productivity * 8,78 0,88

EntireSample

Value added (constant price) 1 308 348 5 606 703Intermediate consumption (constant price) 3 384 245 15 900 000Production (constant price) 4 683 511 20 900 000Number of permanent employees 111 198Total number of employees (permanent and seasonal) 125 248Capital Stock (constant price) 2 609 306 15 600 000Investment (constant price) 345 758 1 499 262Labour Productivity * 8,69 0,90

* This is the unweighted average, on all 11 years, expressed in log.

158


Table 10 shows unweighted average productivity by sector over the

11 years covered in the sample. Throughout the entire period under

consideration, 9 sectors have an average productivity above that of

all businesses. The most productive among these sectors are

Chemicals and Pharmaceuticals (with an unweighted average of

9.49), followed by Electrical Machinery (9.11) and Rubber and Plastics

(9.06). Only 3 sectors have an average productivity below the

productivity of all businesses in the sample. These are Apparel (with

an average productivity of 8.25), Footwear and Leather (8.45), and

Furniture (8.60).

Table 10: Unweighted Average of Labour Productivity by Sector

Sector Unweighted Average of Labour Productivity

1 Agri-food and Tobacco 8.88

2 Textile 8.72

3 Apparel 8.25

4 Footwear and Leather 8.45

5 Timber, Paper & Publishing 8.89

6 Chemicals and Pharmaceuticals 9.49*

7 Rubber and Plastics 9.06

8 Non-metal Materials 8.94

9 Metal Materials 8.77

10 Electrical Machinery, Machines and Appliances 9.11

11 Automotive Industry and other Transportation Equipment 8.85

12 Furniture 8.60*

Table 11: Average Labour Productivity by Capital Structure

Corporate capital structure Unweighted Average Labour Productivity

Businesses that have at least part of their capital held by the State 9.22

Businesses that have at least part of their capital held by foreign private investors 8.71

Table 11 shows unweighted average productivity by corporate

capital structure. It is found that businesses whose capital is held

in whole or in part by the State or foreign investors have, throughout

the period, an average productivity higher than all businesses.

However, no causal link may be inferred, given especially that, as

shown above, these firms fall mainly under the "large" category of

businesses. It is therefore not surprising to observe a higher average

productivity for both categories of businesses.

Hereafter is the descriptive analysis of the labour productivity trend.

Table 12 provides the weighted average productivity per year

(expressed in log), which is also shown graphically (Graph 1).

During these 11 years, the labour productivity of Tunisian

businesses in our sample rose sharply. Labour productivity

(weighted average) increased from 9.42 in 1997 to 9.67 in 2006

(which is a 25% increase) and 9.91 in 2007 (i.e. 49% increase, still

with respect to 1997). With respect to annual growth rates,

productivity declined only between 2002 and 2003 (5%), between

2003 and 2004 (1%) and between 2004 and 2005 (1%). The strong

productivity growth registered between 2006 and 2007 (+24%) is

quite surprising and should be considered with caution. Indeed,

159

the year 2007 is characterized by a significant turnover of

businesses in the sample. As shown in Table 13, 30% of firms in

2007 were never previously present in the database. It seems that

these arrivals and disappearances of businesses have greatly

contributed to such increase in productivity between 2006 and

2007. Although INS uses a number of procedures to ensure the

representativeness of the samples surveyed, caution must be

exercised in interpreting results when working on databases which

are not from censuses. Furthermore, to avoid distorting the

interpretations, some graphs will be presented: (i) covering the

entire period (i.e. from 1997 to 2007) and (ii) leaving out the last

year (i.e. from 1997 to 2006).


Table 12: Labour Productivity Trend for all Businesses in the Sample

YearsWeighted average productivity (in log)

Annual growth rate of weighted averageproductivity

Productivity growth rate with respectto 97

1997 9.42

1998 9.45 3% 3%

1999 9.49 4% 7%

2000 9.58 9% 16%

2001 9.61 3% 19%

2002 9.66 5% 24%

2003 9.61 -5% 19%

2004 9.60 -1% 18%

2005 9.59 -1% 17%

2006 9.67 8% 25%

2007 9.91 24% 49%

Graph 1: Weighted Average of Labour Productivity Trend between 1997 and 2007 (in log)

160


Table 13: Number of Years of Presence of Businesses in the Database by Year

No. Of years of presence

1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007Distribution in 2007

1 131 111 46 74 47 93 127 96 156 235 253 30%

2 83 116 52 113 121 95 90 94 95 166 97 8%

3 134 129 131 140 130 95 84 126 164 159 115 10%

4 122 133 151 172 125 91 93 113 114 113 81 7%

5 183 195 210 226 231 103 107 97 95 99 64 5%

6 185 194 195 231 251 211 139 136 130 143 93 8%

7 115 132 135 157 176 159 144 116 114 115 72 6%

8 107 129 131 162 168 153 137 138 116 109 74 6%

9 115 121 118 131 128 123 122 121 120 87 74 6%

10 116 122 127 118 120 121 121 121 125 121 68 6%

11 89 89 89 89 89 89 89 89 89 89 89 8%

Total 1380 1471 1385 1613 1586 1333 1253 1247 1318 1436 1180 100%

Graphs 2 show the weighted average labour productivity by sector,

first between 1997 and 2007, and then between 1997 and 2006. If

2007 is disregarded, it will be discovered that labour productivity

fell in five sectors: textiles (sector 2), apparel (sector 3), chemicals

and pharmaceuticals (sector 6), rubber and plastics (sector 7) and

automotive (sector 11). For the first 3 sectors (textiles (2) , apparel

(3) and chemicals and pharmaceuticals (6), the strong productivity

growth registered between 2006 and 2007 helps to avoid the

foregoing cuts and to end up in 2007 with productivity levels higher

than in the beginning of the period (i.e. 1997). In Sector 7 (rubber

and plastics), labour productivity also increased sharply between

2006 and 2007, but not enough to exceed the productivity level of

1997. In addition, a glance at the graphs over the period 1997-2007

reveals that only one sector, the automotive sector (sector 11),

experienced a significant drop in its labour productivity. In contrast,

Labour productivity increased in seven sectors: agri-food (sector

1), leather and footwear (sector 4), timber, paper and publishing

(sector 5), non-metal materials (sector 8), metal materials

(sector 9), electrical machinery (sector 10) and furniture (sector 12).

Among these sectors, labour productivity growth is particularly

marked in the food and agriculture sector (1), timber, paper and

publishing (5), non-metal materials (8), metal materials (9) and

electrical machinery (10).

Graphs 3 show the labour productivity trend by size, first between

1997 and 2007 and then between 1997 and 2006. What has been

described, in the graphs, as "product-small", "product-medium" and

"product-large" corresponds to the weighted average labour

productivity, respectively, of "small", "medium" and "large" firms in

the sample. Although, with respect to unweighted average for the

whole period, Table 9 reveals that the labour productivity of "large"

companies exceeded that of "medium" ones, it is found that the

weighted average productivity of these "medium " businesses (red

solid line) grew faster than that of "large" firms (wide dotted, green).

The labour productivity of "medium" enterprises stood at 9.36 in

1997 and increased to 9.91 in 2006 and to 9.51 in 2007. For the

category of "large" companies, the weighted average of labour

productivity stood at 9.42 in 1997. It increased to 9.65 in 2006 and

to 9.91 in 2007. The labour productivity of "medium" firms exceeded

that of "large" firms from 2001 until 2006. In 2007, the productivity

of "large" businesses once more exceeded that of "medium" firms.

The irregular growth of average productivity of "small" businesses

161

(small blue dots) is due most likely to the substantial arrivals and

disappearances from the sample, predominantly by this category

of companies. Moreover, the strong productivity growth between

2006 and 2007 was recorded mainly by "small" firms and, to a lesser

extent, by "large" firms.

Graphs 4 shows the labour productivity trend by corporate capital

structure between 1997 and 2007 and depicts, in solid line, the

category of businesses whose capital is wholly held by the State,

and in dotted line, the category of firms with at least part of their

capital held by foreign investors. It is clear that the productivity

of domestic firms rose more sharply than that of companies

with foreign capital. In 1997, the productivity of domestic

businesses stood at 9.32. It increased to 9.72 in 2006 and to

almost 10 in 2007. For companies with foreign capital, it increased

from 9.58 in 1997 to 9.68 in 2006 and 9.82 in 2007. Between

1997 and 2003, the productivity of companies having foreign

capital is higher than that of domestic firms. Starting 2003, the

reverse is true: labour productivity of domestic firms becomes

higher than that of firms with foreign capital. This mind-boggling

outcome is interesting, and requires a more detailed specific

analysis. Indeed, it is generally expected that businesses owned

in part by foreign investors will experience more substantial

productivity growth.


162


Graphs 2: Trend in Weighted Average of Labour Productivity between 1997 and 2007 (in log)Between 1997 and 2006 (in log)

Entre 1997 et 2006 (en log)

163


Graphs 3: Trend in Weighted Average of Labour Productivity by Sizebetween 1997 and 2007 (in log)

Entre 1997 - 2006 (en log)

Graph 4: Trend in Weighted Average of Labour Productivity by Corporate Capital Structure (in log)

164

2.2 Productivity Decomposition Analysis

The purpose of this section is to identify the source of changes

in productivity. The idea here is, specifically, to know whether

productivity growth arises from increased productivity within

businesses of the sample or ensues from reallocation.

Labour productivity growth in businesses may be related

to either:

- unanticipated cyclical changes in demand by businesses, which

generally account for frequent "unintentional" slumps in labour

productivity;

- labour market rigidities that can slow down the adaptation of

the number of employees (upward or downward) to changes in

production; or

- a set of firm-specific decisions that can lead to improved

productivity. These include, for example, improving the standard

of employee training, investing in the procurement of more efficient

machines, use of better quality inputs, corporate reorganization,

redundancy-related decisions, etc.

The phenomena of reallocation may ensue from inter-sector

changes (some sectors develop while others stagnate or decline),

or intra-sector changes, i.e. market share variations as well corporate

entries and exits occur within each sector. For a long while, because

only sector data (from either domestic or international sources

provided by UNIDO) were available, the reallocation analysis focused

on inter-sector changes, given that the homogeneous firm

assumption posited by both traditional international trade theories

and the New Trade theory (Krugman, 1979, Helpman and Krugman,

1987) does not help to explain, theoretically, the possibility of intra-

industry reallocation. The recent development of the "New New

Trade" theory, initiated in particular by Melitz (2003), and characterized

by consideration of the heterogeneity of businesses within sectors,

justified theoretically, that the analysis should focus on changes

within firms. Access to the individual databases of companies helped

to develop empirical analyses in furtherance of these theoretical

advances.

Hence, the main lessons learned from recent theoretical and empirical

developments in the literature are only within the same industry.

There are companies that can be very outstanding, owing to their

size, degree of integration into the international economy, level of

productivity, etc. and that, in this context, any change (trade reform,

business environment, change in international demand, increased

competition, etc..), will impact differentially on these businesses and

necessarily engender reallocations within sectors. The predominant

concept from the literature is that these intra-industry reallocations

would be of much greater magnitude than those occurring between

sectors. In this theoretical framework, Melitz (2003, cited above) has

shown, for instance, that opening up to international trade leads to

increased market shares for businesses that were initially the most

productive to the detriment of the less productive ones, which

disappear or see their market share shrink. For these authors of

"New New Trade", changes in the aggregate productivity of an

economy are due mainly to the reallocation of such phenomena

within industries, especially when it comes to savings open to

international trade.

Upon its accession to the WTO in March 1995 and the entry into

force of a number of trade agreements, both within the multilateral

framework (GAFTA, which took effect in January 1998, the Agadir

Process, the AMU) and on a bilateral basis (the Euro-Mediterranean

Association Agreement, implemented starting January 1996, the

series of agreements with Egypt, Morocco and Jordan, all three

implemented in 1999), Tunisia undertook to eliminate these customs

duties. Between 1995 and 2004, the simple average of tariffs fell by

22%. Consequently, one can therefore wonder, as recent theoretical

developments would suggest, whether this openness to international

trade has resulted in the huge, concomitant productivity gains from

the phenomenon of reallocation.

2.2.1. Principle of Decomposition Methods

In the literature, the three main methods used are those of Foster,

Haltiwanger Krizan (FHK, 1998 and 2001), Griliches and Regev (GR,

1995) and, more recently, Pavcnik (2002). Although the FHK method

is the most comprehensive, it requires, as does also the GR method,


165


knowledge of the arrival and disappearance of businesses. Given

that data on Tunisian firms do not allow for the identification of "real"

arrivals and disappearances , the only method applicable in this case

is Pavcnik's.

This method consists in decomposing aggregate productivity in the

following manner:

(1)

with , the weighted average of productivity (expressed in log) in

year t,

(2)

, unweighted average of productivity in year t, , market share

of business i in sample, , average market share of businesses in

the sample and , productivity of business i, expressed in log.

The employment or production criterion may be used in obtaining

the market share. In this research work, production26 was chosen.

This decomposition may be carried out either through labour

productivity or through total factor productivity. In this case, only

labour productivity was used.

This decomposition shows that the weighted average productivity

can be decomposed into two terms:

- The first term (i.e.) ) is the unweighted average of productivity

of all firms in the sample. This first term is what is known in the

literature as a "within" effect. It measures the contribution of

business productivity growth.

- The second term (i.e. ) which is called

the covariance term, takes into account the difference in the

company’s market share relative to the average share of firms

in the sample and the productivity gap of the firm with respect

to the unweighted average productivity of the sample. Hence,

the latter term determines the contribution to the productivity

growth of the sample ensuing from the reallocation of market

shares among firms of different productivity levels. The

contribution of this reallocation effect is particularly crucial, given

that companies with relatively high (/ low) productivity, i.e. above

(/ below) the unweighted average of the sample, have market

shares relatively bigger (/ smaller) than (i.e. above (/ below) the

average market share of firms in the sample).

If the first term is positive, it means that, on average, companies

have increased their productivity. If the second term is positive, it

indicates that a higher proportion of goods is produced by more

efficient firms. Although the corporate data used in this study are

from a sample, one might expect, according to the predictions of

theoretical literature, that against the backdrop of liberalization, the

second term is positive and it increases over time, over the period

considered.

2.2.2. Outputs

This decomposition method was applied (i) across the entire

sample, (ii) by sector, (iii) by size and (iv) by corporate capital

structure. In all cases, the results are shown in terms of change

from the start year, i.e. 1997. In the 4 tables below, the second

column indicates changes in aggregate productivity with respect

to 1997. The following 2 columns correspond to variations of the

first and second term in the decomposition. As required by the

decomposition equation, the sum, online, of columns (3) and (4)

corresponds to column (2).

26 By applying this method on data from the UK, Disney et al. (2003) shows that the use of production, alternately with employment, only changes the results very marginally.

166

Table 14 shows the outcome of the entire sample. Column (2),

which indicates changes in aggregate labour productivity for all

businesses, is the last column of Table 12 already presented above.

It is found that much of the productivity growth is derived from

the reallocation effect. In 2006, the 25% aggregate labour

productivity growth rate were due to the 8% from productivity

growth within businesses, and 17% from the reallocation of

resources from less efficient to the most efficient firms. In other

words, 67% of the variation in aggregate productivity over 10 years

(97-2006) is due to the increase in the covariance term. In 2007,

the same share stood at 72%. Although this covariance term did

not increase regularly throughout the period, it is always positive

(except only for the first two years), which shows that reallocation

plays in the right direction, i.e. the most productive firms are

developing and/or the least productive ones have decreasing

market shares.


Table 14: Decomposition of Aggregate Productivity Growth for the Entire Sample

Years Aggregate productivity growthVariation in unweighted Productivity (First term)

Variation inCovariance(Second term)

1997 0.000 0.000 0.000

1998 0.023 0.055 -0.031

1999 0.071 0.072 -0.001

2000 0.153 -0.038 0.191

2001 0.183 0.043 0.140

2002 0.235 0.124 0.112

2003 0.183 0.126 0.057

2004 0.172 0.109 0.062

2005 0.164 0.079 0.085

2006 0.249 0.081 0.168

2007 0.486 0.138 0.348

167


Table 15: Decomposition of Aggregate Productivity Growth by Sector

Sectors Years Agr.Pdt Unwghtd.Pdt

Cov. Sectors Years Agr.Pdt Unwghtd.Pdt

Cov.

1 1997 0.000 0.000 0.000 7 1997 0.000 0.000 0.000

Agri-food &Tobacco

1998 0.043 0.063 -0.020

Rubber andPlastics

1998 0.155 0.197 -0.043

1999 0.044 0.007 0.037 1999 0.226 0.183 -0.032

2000 0.100 0.036 0.063 2000 0.226 0.134 0.092

2001 0.224 0.147 0.077 2001 0.127 -0.017 0.144

2002 0.366 0.318 0.048 2002 0.300 0.419 -0.119

2003 0.456 0.306 0.150 2003 -0.186 0.180 -0.366

2004 0.387 0.396 -0.010 2004 -0.118 0.149 -0.267

2005 0.261 0.219 0.041 2005 -0.311 0.014 -0.326

2006 0.304 0.077 0.227 2006 -0.301 0.076 -0.376

2007 0.303 0.066 0.237 2007 -0.120 0.120 -0.240

2 1997 0.000 0.000 0.000 8 1997 0.000 0.000 0.000

Textile

1998 -0.172 -0.094 -0.078

Non-metalMaterials

1998 0.037 0.151 -0.114

1999 -0.150 0.033 -0.183 1999 -0.085 0.089 -0.175

2000 0.172 0.101 0.072 2000 0.254 0.130 0.125

2001 0.266 0.187 0.080 2001 0.397 0.354 0.043

2002 0.281 0.424 -0.143 2002 0.552 0.400 0.152

2003 0.168 0.277 -0.109 2003 0.527 0.556 -0.029

2004 0.099 0.275 -0.176 2004 0.509 0.653 -0.144

2005 -0.008 0.205 -0.213 2005 0.407 0.562 -0.155

2006 -0.116 0.100 -0.216 2006 0.565 0.585 -0.019

2007 0.164 0.100 0.044 2007 0.700 0.608 0.092

3 1997 0.000 0.000 0.000 9 1997 0.000 0.000 0.000

Agri-food &Tobacco

1998 0.052 0.091 -0.039

MetalMaterials

1998 0.068 -0.053 0.121

1999 0.240 0.093 0.146 1999 0.281 0.057 0.224

2000 0.244 -0.077 0.321 2000 0.236 -0.124 0.360

2001 0.294 0.056 0.237 2001 0.220 -0.215 0.435

2002 0.251 0.100 0.150 2002 -0.028 -0.164 0.136

2003 -0.061 0.111 -0.172 2003 0.067 -0.168 0.235

2004 -0.159 0.022 -0.181 2004 0.198 -0.055 0.253

2005 -0.247 -0.033 -0.213 2005 0.527 -0.166 0.693

2006 -0.076 -0.004 -0.072 2006 0.742 -0.311 1.052

2007 0.654 -0.006 0.661 2007 1.236 -0.257 1.493

168


4 1997 0.000 0.000 0.000 10 1997 0.000 0.000 0.000

Footwear andLeather

1998 0.220 0.034 0.186

ElectricalMachineryand appliances

1998 -0.094 0.016 -0.110

1999 0.250 0.101 0.148 1999 0.117 0.156 -0.039

2000 0.211 0.006 0.206 2000 0.269 0.165 0.104

2001 0.209 0.028 0.180 2001 0.267 0.232 0.035

2002 0.023 -0.120 0.144 2002 0.287 0.249 0.038

2003 0.379 0.036 0.343 2003 0.163 0.374 -0.211

2004 0.226 0.016 0.210 2004 0.345 0.359 -0.015

2005 0.071 -0.027 0.098 2005 0.356 0.425 -0.069

2006 0.302 0.004 0.298 2006 0.469 0.364 0.105

2007 0.333 0.157 0.176 2007 0.504 0.636 -0.132

5 1997 0.000 0.000 0.000 11 1997 0.000 0.000 0.000

Timber, Paper& Publishing

1998 0.031 0.042 -0.012

Automotive Industryand otherTransportationEquipment

1998 -0.131 -0.060 -0.071

1999 0.081 0.057 0.024 1999 -0.370 -0.091 -0.279

2000 -0.014 0.012 -0.026 2000 -0.082 0.028 -0.110

2001 0.012 0.156 -0.143 2001 -0.090 0.048 -0.138

2002 0.045 0.113 -0.068 2002 -0.216 0.074 -0.290

2003 0.177 0.086 0.091 2003 -0.297 -0.300 0.003

2004 0.164 0.172 -0.008 2004 -0.289 0.087 -0.376

2005 0.278 0.102 0.176 2005 -0.332 0.041 -0.372

2006 0.209 0.148 0.062 2006 -0.454 -0.233 -0.221

2007 0.398 0.251 0.147 2007 -0.190 -0.069 -0.121

6 1997 0.000 0.000 0.000 12 1997 0.000 0.000 0.000

Chemicalsand Pharma-ceuticals

1998 -0.051 -0.079 0.028

Furniture

1998 0.064 0.135 -0.071

1999 -0.020 -0.019 -0.001 1999 0.069 0.108 -0.040

2000 -0.326 0.100 -0.426 2000 0.230 0.118 0.112

2001 -0.005 0.237 -0.241 2001 0.273 0.194 0.079

2002 -0.382 0.264 -0.645 2002 0.322 0.348 -0.027

2003 -0.847 0.271 -1.118 2003 0.008 0.018 -0.010

2004 -0.094 0.414 -0.509 2004 -0.007 0.064 -0.071

2005 -0.358 0.272 -0.630 2005 0.208 0.263 -0.055

2006 -0.685 0.153 -0.838 2006 0.217 0.335 -0.118

2007 0.173 0.224 -0.051 2007 0.521 0.513 0.007

Table 15 contains the outcomes of labour productivity decomposition

by sector. While it is true that, for the entire sample, reallocation

contributed significantly to aggregate productivity growth, it is worth

underscoring also that such assertion still needs to be verified in

169


the sectors. In fact, it is only in 2 industries (footwear and leather

(sector 4) and metal materials (sector 9)), that changes in the

covariance term are always positive throughout the period and

higher than corporate productivity. However, in the following six

sectors: textiles (sector 2), chemicals and pharmaceuticals (sector

6), rubber and plastics (sector7), non-metal materials (sector 8),

electrical machinery (sector 10) and furniture (sector 12), corporate

productivity increased, while the covariance term had a negative

impact on the variation of aggregate productivity. In the timber,

paper and publishing sector (sector 5), the dominant effect is labour

productivity growth within businesses. In the agri-food sector

(sector 1), productivity growth is also due, throughout the period,

to the productivity growth of businesses, except for the last 2 years

(2006 and 2007) during which contributions from the covariance

term were particularly significant. Lastly, in two sectors, the apparel

(sector 3) and automotive (sector 11), these two terms (corporate

productivity and covariance), played a negative role on the change

in aggregate labour productivity.

The outcomes of decomposition by size are presented in Table 16.

The impact of reallocation contributed significantly to the aggregate

labour productivity growth for the "average" and "large" categories

of businesses. With the exception of 1998, the variation of the

covariance term was indeed always positive for these two groups

of firms. For the "small" category, that term varied positively only

for four years (2000, 2001, 2005 and 2007). The strong growth

of the covariance term in 2007 should be considered with caution,

given, as already underscored above, the crucial survey sample

rotation particularly relevant to "small" businesses. These results

also show that "medium"-sized businesses, for the most part,

increased their unweighted labour productivity. It would be

, interesting to understand the factors that prompted them to im-

prove their efficiency and the means by which they achieved it.

Table 17 shows the results of the decomposition of aggregate

labour productivity by corporate capital structure. It is fascinating

to observe that, in the sample, the reallocation effect tended to

contribute to aggregate productivity growth only for businesses

that are entirely domestic. For firms with part of their capital held

by foreign investors, the variation in the covariance term is positive

only for 4 years (2000, 2001, 2002 and 2007).

With regard specifically to entirely domestic businesses, the 40%

increase in aggregate labour productivity in 2006 can be broken

down as follows: 7% accounts for labour productivity growth within

businesses and 33% is derived from the reallocation effect. In 2007,

the 68% increase in aggregate productivity ensued from productivity

growth within businesses (14%) and the reallocation effect (54%).

Table 16: Decomposition of Aggregate Productivity Growth by Size

Size YearsVariation inAggregate Productivity

Variation inUnweighted Productivity (First term)

Variation inCovariance(Second term)

Small

1997 0.000 0.000 0.000

1998 -0.081 0.032 -0.113

1999 -0.679 0.057 -0.735

2000 0.091 -0.013 0.104

2001 0.343 0.065 0.278

2002 -0.615 0.031 -0.646

2003 -0.483 -0.039 -0.444

2004 -0.295 -0.059 -0.236

2005 0.248 0.052 0.196

2006 -0.554 0.018 -0.572

2007 1.581 0.154 1.427

170


Size YearsVariation inAggregate Productivity


Variation inCovariance (Second term)

Medium

1997 0.000 0.000 0.000

1998 0.064 0.081 -0.018

1999 0.125 0.077 0.049

2000 0.204 -0.049 0.253

2001 0.218 0.046 0.173

2002 0.377 0.200 0.177

2003 0.410 0.251 0.159

2004 0.378 0.205 0.174

2005 0.278 0.170 0.108

2006 0.560 0.161 0.399

2007 0.147 0.143 0.004

Large

1997 0.000 0.000 0.000

1998 0.022 0.048 -0.026

1999 0.088 0.060 0.028

2000 0.151 -0.063 0.214

2001 0.176 -0.007 0.183

2002 0.232 0.040 0.192

2003 0.171 0.064 0.106

2004 0.157 0.023 0.134

2005 0.152 -0.036 0.188

2006 0.225 0.024 0.201

2007 0.488 0.095 0.393

Table 17: Decomposition of Aggregate Productivity by Corporate Capital Structure

Capital Structure

YearsVariation inAggregate Productivity


Variation inCovariance (Second term)

DomesticFirms

1997 0.000 0.000 0.000

1998 0.095 0.054 0.041

1999 0.102 0.063 0.038

2000 0.150 -0.032 0.183

2001 0.241 0.049 0.192

2002 0.285 0.136 0.150

2003 0.287 0.116 0.171

2004 0.325 0.122 0.203

2005 0.388 0.094 0.294

2006 0.399 0.073 0.326

2007 0.681 0.144 0.537

Firms withforeign capital

1997 0.000 0.000 0.0001998 -0.086 0.055 -0.1411999 0.016 0.096 -0.0802000 0.127 -0.050 0.1772001 0.079 0.031 0.0482002 0.137 0.101 0.0372003 0.024 0.141 -0.1172004 -0.044 0.087 -0.1312005 -0.132 0.053 -0.1862006 0.040 0.095 -0.0552007 0.236 0.126 0.110

171


3. Conclusion

In this study, the labour productivity of Tunisia’s industrial sectorcompanies between 1997 and 2007 was analysed using a sample

of individual firms from annual surveys. The key results of this

analysis are as follows:

First, the aggregate labour productivity of Tunisian businesses

rose sharply. It increased by 25% between 1997 and 2006 (and

by 49% between 1997 and 2007, although the past year should

be considered with extreme caution, given that 30% of the sample

was renewed);

Second, at the sector level, aggregate labour productivity

increased in seven industries (agri-food, leather and footwear,

timber, paper and publishing, non-metal materials, metal materials,

electrical machinery and furniture). However, if year 2007 is

disregarded, aggregate productivity fell in five sectors (textiles,

apparel, chemicals and pharmaceuticals, rubber, plastics and

automotive);

Third, while unweighted average productivity throughout the period

is higher for "large" firms than for "average" ones, aggregate


"large" firms. From 2003 to 2006, the aggregate labour productivity

of "average" businesses exceeds that of "large ones;


more sharply than that of companies with at least part of their capital

held by foreign investors; and





accounted for by 8% in labour productivity growth within businesses

and 17% from the reallocation effect. This is true especially for

domestic firms and "medium-" and "large-"sized firms. However,

at the sector level, this result is only true for two industries (footwear

and leather, metal materials). Labour productivity growth within

businesses involved a larger number of sectors (food and

agriculture, textiles, timber, paper and publishing, chemicals and

pharmaceuticals, rubber and plastics, non-metal materials, electrical

machinery and furniture).

172


Bibliography

Disney, R., J. Haskel and Y. Heden (2003), Restructuring and

productivity growth in UK manufacturing, Economic Journal, Vol.

113, No. 489, pp. 666 – 694.


productivity growth: Lessons from microeconomic evidence, Working

Paper 6803 NBER.


productivity growth: Lessons from microeconomic evidence, in

Edward Dean, Michael Harper, and Charles Hulten (eds.), New

Developments in Productivity Analysis, Chicago: University of

Chicago Press.

Griliches, Z. and H. Regev (1995), Firm productivity in Israeli industry

1979-1988, Journal of Econometrics, 65, pp. 175-203.

Hall, B. H. and J. Mairesse (1995), Exploring the relationship between

R&D and productivity in French manufacturing firms, Journal of

Econometrics, Elsevier, 65(1), pp. 263-293.

Helpman E. and P.R. Krugman (1987), Market Structure and

Foreign Trade: Increasing Returns, Imperfect Competition, and the

International Economy, MIT Press Books, The MIT Press.

Krugman, P. R. (1979), Increasing returns, monopolistic competition,

and international trade, Journal of International Economics, 9(4),

pp. 469-479.

Melitz, M. (2003), The Impact of Trade on Intra-Industry

Reallocations and Aggregate Industry Productivity, Econometrica,

71, pp. 1695–1725.

Pavcnik, N. (2002), Trade Liberalization, Exit, and Productivity

Improvements: Evidence From Chilean Plants, Review of Economic

Studies, 69, pp.245-276.

Modeling and Analysis of Tunisia's Productive System...32 2.3.2 Global Demand Trend for Tunisian...

Documents

Transcript of Modeling and Analysis of Tunisia's Productive System...32 2.3.2 Global Demand Trend for Tunisian...